Layered representations for image coding
Edward H. Adelson
Media Laboratory
Massachusetts Institute of Technology, Cambridge, MA 02139
e-mail: adelson@https://www.doczj.com/doc/7212613874.html,
1.0 Introduction
In order to represent a television image sequence with a minimal number of bits, it is nec-essary to perform ef?cient image coding. Transform coding, subband coding, and vector quantization are examples of popular techniques that have been applied to the coding of still and moving images. In the case of moving image sequences, these techniques are often combined with frame-differencing or motion compensation, to take advantage of the temporal structure of image sequences.
Any image coding technique contains an implicit vocabulary and image model; that is, it offers a way to describe images. For example, in a discrete cosine transform technique, an image is described as a sum of cosine basis functions, and this forms the implicit image model. If motion compensation is used for moving images, then the implicit image model includes the idea that regions of the image move coherently across time.
There are three interrelated parts to the design of an image coding system: the encoding process, the representation, and the decoding process. The choice of representation is the central determinant of the system as a whole: the encoding process converts the input images into the chosen representation, and the decoding process converts the representa-tion into an array of pixels for the display. The representation must be chosen before the encoding or decoding procedure can be designed.
I describe a new image representation based on the concept of "layers" that captures sev-eral important aspects of images and may substantially improve the ef?ciency and ?exi-bility of image coding for many image sequences.
The layered representation also allows for high-quality frame-rate conversion, so that, for example, an original image sequence shot at 50 hz, can be converted to a frame rate of 60 hz without introducing artifacts.
When an image sequence is represented as layers, it also becomes possible to perform interesting editing and post-production effects that would be impossible with ordinary rep-resentations.
2.0 Limitations of existing motion compensation techniques:
In standard motion compensation methods, the contents of one frame is estimated by applying a spatial transformation to the contents of a previous frame. One popular tech-nique is block-matching, in which a given block of pixels in one frame is estimated as the rigidly translated block from a previous frame. If the translated block correctly estimates the corresponding block in the new frame, then it is only necessary to send the translation vector. If the estimate is imperfect, then one must also send an error signal to correct it. In more sophisticated methods, the mapping from one frame to another may involve smooth distortions, including af?ne transformations or rubber-sheet warps.
These techniques are founded on the assumption that the optic ?ow ?eld is smooth and single-valued. This assumption is frequently violated. For example, when a foreground object occludes a background object, the ?ow ?eld is discontinuous at the occlusion boundary, and information at the occlusion boundary disappears while information at the “disocclusion” boundary appears. The occlusion and disocclusion information cannot be captured with the motion compensation, and must be explicitly coded with error signals; this leads to inef?cient coding.
The standard motion assumptions are also violated in the case of transparent motion, where there are two valid motion vectors at the same image location. Transparent motion occurs when, for example, a scene is viewed through a windowpane, and dirt or specular re?ections on the pane are combined with the scene viewed through the pane. Another form of transparency occurs when the edge of a foreground object is blurry, due to optical blur or motion blur, and the blurred edge combines with the background in a transparent manner. There can be two different valid motions in the overlapping blurred region.
3.0 Layered representations
A more sophisticated image model is based on the notion of image layers: the image is considered to consist of a set of overlapping layers, with some ordering in depth. This is the representation used by traditional cel animators, who paint a set of images on transpar-ent plastic sheets and then stack them up to produce the ?nal images. It is, after all, much easier to paint a background once and move the characters in front of it than to repaint the entire scene for every frame. The cels contain areas of varying color and transparency, mimicking the image formation process of objects in the natural world. This is also the model used in many computer animation programs, in which multiple image layers are de?ned, and each layer has its own motion, its own depth, and its own pattern of color and intensity, and its own pattern of transparency.
The concept of layers is used in computerized compositing, as applied in computer graph-ics. This method is well-known (see [1][2]). Figure 1(a) shows a background image,
E0(x,y). Figure 1(b) shows in image of a baseball, E1(x,y), which is to be pasted in front of the background. The pasting is achieved with a compositing mask, A1(x,y) shown in (c). In the simplest appraoch, this is a binary mask, with values 1 and 0, shown as white and black. The background is multiplied by the mask to remove the region hidden by the base-ball, producing the modi?ed background shown in (d). Then the foreground is multiplied by the inverse of the mask, (1-A1(x,y)), to select the baseball itself; this is added to the
background to produce the ?nal image shown in (e), which may be called I1(x,y). Thus the compositing occurs according to the equation:
I1(x,y) = E0(x,y)A1(x,y) + E1(x,y)(1-A1(x,y)). (1)
Figure 1 . (a): The background image. (b) The foreground image. (c) The compositing mask. (d)The background after removal of the foreground region. (e) The ?nal image. (f) The background image.
(g) The foreground image under motion blur. (h) The compositing mask under motion blur. (i) The background attenuated by the compositing mask. (j) The ?nal image.
Insert ?g. 1 here (baseball compositing)
In a more sophisticated approach to compositing, the mask can take on continuous values between 0 and 1. For example, consider the case where the baseball is blurred due to motion. In the blurred edge region, the image consists of an average of the background and the (blurred) baseball. The proportions of background and foreground depend on the distance from the edge. This situation can be captured in the same computational frame-work as before. Again, Figure 1(f) shows the background E0(x,y); (g) shows the fore-ground E1(x,y), which contains motion blur; (h) shows the compositing mask, A1(x,y), which contains the same motion blur; it takes on values between 0 and 1; (i) shows the
modi?ed background, which is attenuated in varying amounts by the mask; and (j) shows the ?nal composited image.
(Note that in computer graphics, the variable α is usually used, with the inverse of the con-vention used here for A. That is, A = (1-α).)
The same processing stages may be applied to combine a series of layers, as shown in Fig-ure 2. The background image is E0. It is combined with the foreground image E1 under the control of the attenuation map A1. The box labeled “inv” computes the inverse mask func-tion, (1-A1). The result is the image I1. This image may then be combined with the next layer, E2, under control of the mask A2, to produce the image I2.
The basic compositing stage is enclosed in the dashed rectangle. One may repeat as many stages as desired.
Figure 2 A ?ow chart for compositing a series of layers.
inv E1
A1
inv E2
A2
E0
I0I1I2
This example illustrates how layers may be used to produce a single static image. The same concepts can be extended to produce a powerful and ?exible method for encoding image sequences.
4.0 Layers for image sequences
Suppose that the baseball in the prior example is translating across the scene. Then it will produce the series of images shown in Figure 3 as (a1) to (a3). In order to synthesize this sequence, we can use the set of images shown in the left column, (b1), (c1), (d1), and (e1). (b1) is the stationary background; (c1) is the baseball image; (d1) is the mask; and (e1) is the velocity map that is applied successively to (c1) and (d1). After one application, the baseball and the mask are as shown in (c2) and (d2); after compositing they produce the image (a2). After a second application they are as shown in (c3) and (d3), leading to the image (a3).
In the proposed encoding method, the original image sequence, (a1) to (a3), could be rep-resented with the set of maps (b1), (c1), (d1), (e1). Thus, rather than explicitly represent-ing the sequence, onerepresents a set of component layers along with rules for their transformation and combination. This approach offers several advantages. The encoding can be ef?cient, since each of the components only has to be encoded once, as a still image, and the change from one frame to the next is captured in the very simple velocity
map. In addition it is possible to adjust the speed of the motion or to adjust the frame rate by changing the amplitude of the velocity map
..
Figure 3 . Frames 1 to 3 form a sequence of images of a moving baseball seen against a stationary background. (a1) to (a3) are the ?nal frames. (b1) is the background. (c1) to (c3) are the set of baseball images, and (d1) to (d3) are the set of masks. (e1) shows the motion map that is applied to the baseball and the corresponding mask in order to generate the image sequence
In the above example, the motion is a simple translation. One could also use an af?ne transformation in order to achieve rotation, dilation, and shearing, or one could use more complex warp maps to achieve arbitrary distortions.
5.0 Delta maps
The velocity map will not always convey all the changes occuring to a given layer. There-fore it is useful to add information from a “delta map” (change map). This is an image that de?nes the temporal derivative at each point in a layer. Thus, in addition to undergoing a warp from one frame to the next, each layer can be combined with an additive correction.
As before, this change map can be scaled in order to interpolate or extrapolate to different points in time. (A multiplicative change map may also be used to account for temporal variations in contrast.))
A combined system incorporating velocity maps, V(x,y), and delta maps, D(x,y), is shown in Figure 4. From one frame to the next, the background E 0 is modi?ed by the delta map D 0 and then warped by the velocity map V 0 . This is then composited with the image E 1 , which is itself added to D 1 and warped by V 1 . The attenuation mask, A 1 , is warped by the same velocity map, V 1 .
Figure 4 . A ?ow chart for compositing that incorporates velocity maps, V , and delta maps, D.
6.0 Comparison with conventional techniques.
It is illuminating to compare the layered approach with the more conventional approaches to coding image sequences. In Figure 5 the three frames are shown as (a1) to (a3). The tra-inv
E 1A 1
I 0I 1
warp
V 1D 1
E 0
warp V 0D 0warp
ditional frame-difference method is shown in the middle row. First an initial frame is coded, shown in (b1); then the frame differences are coded for each succeeding frame, shown in (b2) and (b3).
The traditional motion-compensated method is shown in the bottom two rows. First the initial frame is coded, as in (c1); then the blocks shown in (c2) and (c3) are used to show regions that are considered to be in rigid translation. Finally the error images, (d2) and (d3), are used to correct for the failures of motion compensation.
Figure 5 (a1) to (a3): original image sequence. (b1): initial frame for frame-differencing. (b2), (b3): difference frames. (c1): initial frame for motion compensation. (c2), (c3): moving image blocks for motion compensation. (d2), (d3): error signals added to correct for motion compensation.
Neither the frame-differencing method nor the motion-compensated method uses a very accurate model of image formation, and for this reason the amount of corrective informa-tion that must be sent is quite large. In many cases, the layered format will therefore be more ef?cient. The layered format will have its greatest advantage in cases where the lay-ered model accurately represents the image information over many frames, so that one can send a minimum of corrective information in the form of change maps.
One should also note that neither the frame-differencing nor the motion-compensated approach allow for good temporal interpolation. It is true that one can add partial amounts of the corrective signals, in an attempt to synthesize an intermediate frame, but in a tradi-tional motion-compensated technique this will lead to poor results due to “ghosting.” The layered format can prevent these artifacts.
7.0 Specifying the layered format:
In a digital system we can attach a rich variety of image information to the layers. In addi-tion to color and transparency, we can attach information about motion, blur, and depth. An example of such a format is as follows:
There are N layers, which are typically ordered in depth. Associated with each layer is a set of maps. These maps are de?ned at a set of discrete 2-D locations, (x,y), and a set of times, t. The locations would normally be the spatial sampling lattice and the times would normally be the frame times. Note, however, that it is not necessary that all of the maps be sampled on the same lattice in space and time. As long as interpolation rules are given, it is possible to synthesize the requisite values of the maps at any position and time. Thus the different maps may be sampled on different lattices in space and in time. (Indeed there is no need to de?ne the map by a set of samples; it could also be de?ned by a function or a set of coef?cient).
The basic maps are as follows:
E(x,y,t), the intensity map (where E may be a vector such as (R,G,B) in the case of color images).
A(x,y,t), the attenuation map, where A is between 0 and 1. Each layer multiplies the image intensities below it by A, and adds its own intensity E, scaled by (1-A).
D(x,y,t), a “delta map” (change map) that describes the temporal derivative of the points in E(x,y,t). This can also serve as an error signal to correct for failures of the other tempo-ral variation (motion) to adequately describe the signal.
V(x,y,t), a velocity map, which tells how all points in the layer are warped over time. In some cases it will be possible to represent this map with a few parameters, as in the case of translation, or af?ne distortion.
In addition, the following optional information may be included:
C(x,y,t), a contrast change map, which speci?es how the intensity map, E(x,y), should be multiplied to make the transition from one time to the next.
B(x,y,t), a blur map, which can be used to add motion-blur or focus-blur to each point in E(x,y,t). The blur map contains a set of parameters for a space-variant ?lter that is to be applied at each point in the image. In the simplest case the ?lter is a ?xed kernel for the entire layer.
Z(x,y,t), a depth map.
N(x,y,t), a surface orientation map, coding the surface normals (also known as a “bump map”).
For ef?ciency one may wish to associate a bounding box with each layer as well; outside of the box all the values revert to the default, e.g. E=0; A=1; D=0; V=0, etc.
8.0 Temporal interpolation:
If a different set of maps are de?ned for every layer and every frame in time, then there might seem to be no need for the time dependent maps such as V(x,y,t), and D(x,y,t). But these maps have two uses. One is to do temporal interpolation between frames; the other is to allow ef?cient representation.
The virtue of temporal interpolation is this: if the image sequence is represented at one frame rate, but one wishes to display it an a different frame rate, then it is possible to syn-thesize intermediate frames. Given a velocity map and a delta map it is easy to smoothly interpolate to any intermediate time.
In addition these maps may be used for image data compression. It might be possible, for example, to send a sequence of 4 images with only a single intensity map for each layer, along with a single velocity map for each layer.
9.0 Special cases:
There are many phenomena in image sequences that lend themselves naturally to the lay-ered representation. A shadow moving over a surface can be represented as a purely multi-plicative layer, that is, a layer with values for A(x,y), but zero values for E(x,y). Conversely, a specular re?ection such as the re?ection in a window pane is a purely addi-tive layer, with non-zero values for E(x,y) and a constant value for A(x,y). A change in contrast (such as occurs in a fade or with changes in lighting) can be expressed in terms of a layer with zero-valued E(x,y), and a time-varying A(x,y). It can also be expressed through a contrast change map, C(x,y).
10.0 Editing and special effects:
There are many interesting special effects that can be performed if an image has been con-verted into a layered format. Indeed, one of the most important class of special effects in ?lm and video, known as “blue-screening” or “matting,” involved the creation of layers. In blue-screening, an actor is ?lmed performing in front of a blue screen. Later, all blue pixels are classi?ed as belonging to the background, and all non-blue pixels are classi?ed as belonging to the foreground. It then is possible to paste the image of the actor (the fore-ground) onto another background. This technique is widely used in the television and ?lm industries.
The disadvantage of blue-screening is that it requires an arti?cial controlled environment. It would be prefereable to be able to segregate the actor from the background of any scene. Conversion of the image sequence into a layered format would allow one to manipulate existing footage. (Ideas along these lines are being explored by the Movies of the Future Group at the MIT Media Lab).
The layered format would also allow one to manipulate the motions of objects in the scene, so as to speed them up or slow them down, or change their trajectories. It would also allow one to synthetically add or remove objects from a scene, as is done now with blue-screening. It would also be a convenient format for combining natural image sequences with and synthetic image sequences generated by computer graphics.
11.0 The encoding process:
The layered representation and the procedure for decoding it are well-de?ned, and they constitute the core of the approach described here. But the encoding procedure is a sepa-rate problem and it is extremely challenging. The encoding process is non-unique: there are innumerable ways of representing a given image sequence within the constraints of the format described here. Some of the encodings will be useful and ef?cient; others will not. For example, one could simply use a single layer, in which case the system would revert to standard image representations.
Finding an optimal encoding of a given image sequence requires powerful and sophisti-cated techniques of image analysis and machine vision. Current techniques will produce sub-optimal encodings. However, current techniques can be considered to use special limited versions of the layered representation, which is to say that the layered approach offers a superset that includes current techniques. Thus the layered format can provide a standard that is bakcward compatible with existing techniques, but which allows for the growth toward more sophisticated techniques, as the technology for doing image analysis improves.
Motion analysis based on the methods described by [4][5][6][7][8] may be useful. Single-frame layering techniques as described by [3][10][11][12] etc. may be useful.
I describe here one example of a procedure for converting an image sequence into a lay-ered format. This technique relies on various assumptions that will not hold in the general case, but it serves as a useful illustration of how one might procede.
Suppose that a foreground object, with emittance F(x,y,t) moves in front of a background with emittance G(x,y,t) , and that they are combined with an attenuation map A(x,y,t) . The observed image is:
Now suppose that the foreground is transformed, from frame to frame, by a warp operator (velocity map) P , and that the foreground is transformed by Q
, so that
which may be expressed in the simpli?ed notation,
The attenuation mask is transformed in parallel with the foreground, so that
The sequence is then:
I x y t ,,()A x y t ,,()G x y t ,,()1A x y t ,,()–()F x y t ,,()
+=F x y t ,,()P t F x y 0,,()()
=G x y t ,,()Q t G x y 0,,()()
=F t P t F o
=G t Q t G o
=A t P t A 0
=A x y t ,,()P t A x y 0,,()()
=I t A t G t 1A t –()F t
+=
We can then stabilize the foreground and the attenutation map by applying an inverse warp, P -1, leading to:
Now if we de?ne
then we have,
Now suppose that we know the warps P and Q , and we know the background G 0. Then we can determine A 0 as follows. For two different frames, taken at times t1 and t2, we have:
and thus,
Thus, A 0 can be determined from the known quantities on the right hand side. And once A 0 is known, one can determine F 0 by plugging A 0 back into the original equation.
Since the solution will be unstable when the denominator is near zero, it will be advanta-geous to make multiple observations at multiple times and to take a least squares solution:
P t 1–I t A 0P t 1
–G t 1A 0–()F 0+=G t ?P t 1–Q t G 0=I t ?P t 1
–I t
=I t ?A 0G t ?1A 0–()F 0
+=I t 1?I t 2?–A 0G t 1?G t 2?–()
=A 0I t 1?I t 2?–G t 1?G t 2
?–----------------------------=A 0I n ?I m ?–()G n ?G m ?–()
n
m ,∑G n ?G m ?–()
2n m ,∑-----------------------------------------------------------------------=
12.0 Image data accumulation.
The information stored in a layer can extend over a larger visual angle than is presented on a given image. Consider the case where a camera pans over a stationary scene. Each image presents a glimpse into this scene through a rectangular window. To represent this image sequence, one would like to accumulate all of the information into a single large image; later, the whole sequence can be replayed by simply specifying the portion of the scene that is to be presented on each frame. The concept is illustrated in Figure 6.
When information is accumulated over frames, one must also consider how the individual samples in the frames are to be placed into the individual sample points in the accumu-lated layer, as shown in Figure 7. The layer will have a sampling lattice of a certain resolu-tion, and the incoming images will generally present samples that do not align precisely
Figure 6 : Top: the frames in the sequence are take from an original scene that is larger than any individual frame. Bottom: the information from all the frames may be accumulated into a single large layer. Each frame is then a glimpse into this layer as viewed through a window.Original scene
I1I2
I3I4I5I6I7
Accumulated layer
Visible portion
at one instant.
Individual frames
with those of the layer. In this case it is advantageous to think of the underlying layer as a continuous function that is represented by a set of samples, and to think of the incoming data as having been acquired from a continuous signal by some sampling process (which is dependent on the pre?ltering characteristics of the camera). Then one can de?ne a method for estimating the correct layer sample values given the set of observed image samples. It will often be possible to generate a layer with higher spatial resolution than any of the individual images, especially when the camera system is spatially undersam-pled (as in interlace), or has some degree of aliasing (as in a CCD array). (Work along these lines was done at the Media Lab by Suenaga[15]). For display, one takes samples from the layer, but since the samples my fall on intermediate positions, it will again be necessary to use a form of interpolation. Popular methods such a bicubic interpolation can be used.
In some cases, different images in a sequence will have different degrees of resolution. One would prefer to retain the highest resolution at each point. In this case the information can be combined by the method described by Adelson[9]. A Laplacian pyramid is built from each image, and the pyramid coef?cients from the different images are compared. The coef?cient with the largest deviation from zero is retained. Adelson described this
Figure 7 : The layer samples are represented on the rectangular lattice. The incoming samples from frames “1”, “2” and “3” may fall at intermediate positions on the lattice. One may interpolate from these samples in order to generate a best estimate of the underlying layer.
1
1111111
1111111
11111111
1111122222222222222222222222
222223333333
333333333333333333333
method in the context of Laplacian pyramids; however, it could also be used with subband representations such as pyramids using quadrature mirror ?lters, wavelets, or steerable ?l-ters. It could also be used with non-subsampled representations rather than pyramids. Also, rather than using the coef?cient value as the selection criterion, one could use a local energy measure computed for each position in the subband.
13.0 Data compression:
The layered representation, by itself, does not necessarily offer data compression. To achieve data compression the representation should be used in conjunction with other image coding techniques. Each of the maps used in the layers can be compressed using standard image coding techniques such as transform coding or subband coding. Many of the maps will be greatly compressed in this way. For example, the velocity map will prob-ably be of low spatial resolution, and therefore will be greatly compressed. In some cases the velocity will be a simple warp for an entire layer, and thus can be represented by a few parameters. The attenuation map will often be nearly 1 or 0 most places, and will only take on intermediate values in a few places, so it too can be greatly compressed. In addition, it is possible to encode maps in terms of each other. For example, the blur map can often be computed if one knows the velocity map and the effective shutter duration of the camera, as well as information about the camera's focus and the depth of the layer. In these cases some of the maps can be encoded with only the smallest amount of additional information.
14.0 Depth encoding:
Depth is important mainly because it determines which layer is in front of which. In many cases the ordinal relationships are enough, and one can just assign integers to the layers. It may occasionally be necessary to change the numbering. For instance a new object might enter the scene, in which case it might have to squeeze between, say, layers 1 and 2. Thus the new object is given layer 2 and the old layer 2 is renumbered 3. As long as all the lay-ers are renumbered there should be no problem with this abrupt change.
It may be simpler to assign z values, or even z-maps, to the layers, in which case it is more natural to insert one layer between preexisting layers. The z values would not necessarily be accurate representations of object distances. For example, a still scene would initially be treated as a ?at plane, with uniform and arbitrary z; only as things moved in the scene would the z-map evolve. Rendering is accomplished by looking at the z values of every layer at each pixel, and doing the right combination based on the ordering of the z's. (In the simplest case the values of the z's don't matter except as they affect the ordering).
A single layer may split into two if a stationary object suddenly moves. There will be a threshold rate of movement: slow movements will be encoded as warps of the single layer plus an error map to account for the occlusions and disocclusions; faster movements will demand a split into layers.
The z-map could also be used to help encode focal blur. A z-map can also be used to gen-erate a bump map (its gradient) which may allow ef?cient encoding based on shading and light-source direction. A z-map can also account for non-uniform warps due to camera motion.
15.0 References
[1]Porter, T., and Duff, T, “Compositing Digital Images,” Computer Graphics, v. 18, #3, 253-259 (1984).
[2]Duff, T., “Compisiting 3-D Rendered Images,” Siggraph Proceedings, v. 19, 41-44 (1985)
[3]Adelson, E.,H., and Anandan, P., ``Ordinal characteristics of transparency,'' AAAI Workshop on Qualitative Vision, Boston, MA, pp.77-81 (1990).
[4]Shizawa, M., and Mase, K., “A uni?ed computational theory for motion transparency and motion boundaries based on eigenenergy analysis,” Proc. CVPR, Maui, HI, 289-295 (1991).
[5]Girod, B., and Kuo, D., “Direct estimation of displacement histograms,” Proc. Image Understanding and Machine Vision Workshop, Cape Cod, MA pp73-76 (1989).
[6]Bergen, J., Burt, P., Hingorani, R., and Peleg, S., “Computing two motions from three frames,” Proc. 3rd ICCV, Osaka, Japan, pp. 27-32 (1990).
[7]Darrel, T.and Pentland, A., “Robust estimation of a multi-layered motion representation,” Proc. IEEE Motion Workshop, Princeton, NJ., pp. 173-178 (1991);
[8]Black, M., and Anandan, P., “A model for the detection of motion over time,” Proc. 3rd ICCV, Osaka, Japan (1990).
[9]Adelson, E. H., “Depth of ?eld imaging process method,” U. S. Patent no. 4,661,986 (1987).
[10]Kersten, D., “Transparency and the cooperative computation of scene attributes,” in Landy, J., and Movshon, J., eds., Computational Models of Visual Processing, MIT Press (1991).
[11]Williams, L. R., “Perceptual organization of occluding contours,” Proc. 3rd ICCV, Osaka, Japan, 133-137 (1990).
[12]Nitzberg, M., and Mumford, D., “The 2.1-D sketch” Proc. 3rd ICCV, Osaka, Japan, 138-144 (1990)
[13]Lin, H.-D., and Messerschmitt, D., “Video composition methods and their semantics, IEEE ICASSP (1991).
[14]Metelli, F., “The perception of transparency,” Scienti?c American, v. 230, (4), 91-98 (1974).
[15]Suenaga’s work at the Media Lab on resolution out of time.
注意：答案仅供参考 第一章 一、名词解释 图形；图像；点阵表示法；参数表示法； 二、选择题： 1. 下面哪个不是国际标准化组织（ISO）批准的图形标准。（D ） A.GKS B.PHIGS C.CGM D.DXF 2. 下面哪一项不属于计算机图形学的应用范围？（B） A. 计算机动画； B. 从遥感图像中识别道路等线划数据； C. QuickTime技术； D. 影视三维动画制作 3. 关于计算机图形标准化的论述，哪个是正确的（B ）； A. CGM和CGI是面向图形设备的接口标准； B. GKS、IGES、STEP均是ISO标准； C. IGES和STEP是数据模型和文件格式的标准； D. PHIGS具有模块化的功能结构； 4. 与计算机图形学相关的学科有_ A、C、D___。 A. 图像处理 B. 测量技术 C. 模式识别 D. 计算几何 E. 生命科学 F. 分子生物学 三、判断题： 计算机图形学和图像处理是两个近似互逆的学科。（F） 计算机图形学处理的最基本的图元是线段。（F） 四、简答题： 图形包括哪两方面的要素，在计算机中如何表示它们？ 阐述计算机图形学、数字图像处理和计算机视觉学科间的关系。图形学作为一个学科得以确立的标志性事件是什么？ 试列举出几种图形学的软件标准？工业界事实上的标准有那些？举例说明计算机图形学有哪些应用范围，解决的问题是什么？ 第二章 一、选择题：
1. 触摸屏是一种（C ） A. 输入设备； B. 输出设备； C. 既是输入设备，又是输出设备； D. 两者都不是； 2. 空间球最多能提供（D ）个自由度； A. 一个； B. 三个； C. 五个； D. 六个； 3. 等离子显示器属于（C） A. 随机显示器； B. 光栅扫描显示器； C. 平板显示器； D. 液晶显示器； 4. 对于一个1024×1024存储分辨率的设备来说，当有8个位平面时，显示一帧图像所需要的内存为（A、D） A. 1M字节； B. 8M字节； C. 1M比特； D. 8M比特； 5. 分辨率为1024*1024的显示器，其位平面数为24，则帧缓存的字节数应为（A） A. 3MB； B. 2MB； C. 1MB； D. 512KB； 6. 下面对光栅扫描图形显示器描述正确的是：（A） A. 荧光粉涂层均匀离散分布； B. 是一种点画设备； C. 电子束从顶到底扫描； D. 通过控制电子束的强弱实现色彩的强弱； 7. 一个逻辑输入设备可以对应（C）物理输入设备。 A. 仅一个 B. 仅二个 C. 多个 D. 以上都不是 8. 彩色阴极射线管的三基色指得是：（A、C、D） A. 绿色； B. 黄色； C. 蓝色； D. 红色； 9. 计算机显示设备一般使用的颜色模型是（A） A. RGB B. HSV
第一单元点阵绘图 1.1 1、文件/新建，像素、像素，宽度：640、高度：480，72像素/英寸，RGB模式； 2、调出文件Yps1-01.TIF，选择/全选，编辑/定义图案，回到新建文件，编辑/填充/图案/新图案，文件/存储（保存背景）； 3、正圆选区（ALT+SHIFT键），前景色为纯白色(255,255,255)，背景色为纯红色 (255,0,0)，进行渐变（前景色到背景色渐变、径向渐变），画笔工具（柔角100像素，一笔完成）； 4、多边形套索，所有边为直线段，不缺拐角点，前景色为纯黄色(255,255,0)，编辑/填充/前景色； 5、文本工具（横向选区），字体为Arial，Bold（粗体），尺寸140Pt，字间距为20； 6、前景色为纯绿色(0,0,255)，进行渐变（透明条纹渐变、线形渐变），选择/取消选区； 7、前景色为纯黄色(255,255,0)，编辑/描边（3px、居外）； 8、文件/存储为（保存文件）。 第二单元路径的运用 2.1 1. 文件/新建，像素、像素，宽度：640、高度：480，72像素/英寸，RGB模式； 2 .调出文件Yps2-01.TIF，缩放工具（放大），钢笔工具创建路径（小鸟的腿部和爪部不用绘制）； 3 .用路径组件选择工具将Path全部选中，编辑/拷贝，回到新建文件，编辑/粘贴，复制路径，左侧路径名为Alpha 1，右侧路径名为Alpha 2； 4. 左侧路径：将路径作为选区载入，选择/存储选区（Alpha 1），前景色为纯蓝色(0,0,255)，编辑/填充/前景色，前景色为纯白色(255,255,255)，画笔工具（尖角27像素）； 5. 右侧路径：将路径作为选区载入，选择/存储选区（Alpha 2），前景色为纯黄色(255,255,0)，编辑/填充/前景色，，前景色为纯红色(255,0,0)，画笔工具（尖角27像素），画笔工具（柔角27像素）绘制边缘处，选择/反选，前景色为黑色(0,0,0)，画笔工具（柔角27像素）绘制阴影； 6. 文件/存储（保存文件）。
遥感数字图像处理-要点 1．概论 遥感、遥感过程 遥感图像、遥感数字图像、遥感图像的数据量 遥感图像的数字化、采样和量化 通用遥感数据格式（BSQ、BIL、BIP） 遥感图像的模型：多光谱空间 遥感图像的信息内容： 遥感数字图像处理、遥感数字图像处理的内容 遥感图像的获取方式主要有哪几种? 如何估计一幅遥感图像的存储空间大小? 遥感图像的信息内容包括哪几个方面？ 多光谱空间中，像元点的坐标值的含义是什么？ 与通用图像处理技术比较，遥感数字图像处理有何特点？ 遥感数字图像处理包括那几个环节？各环节的处理目的是什么？ 2.遥感图像的统计特征 2.1图像空间的统计量 灰度直方图：概念、类型、性质、应用 最大值、最小值、均值、方差的意义 2.2多光谱空间的统计特征 均值向量、协方差矩阵、相关系数、相关矩阵的概念及意义波段散点图概念及分析 主要遥感图像的统计特征量的意义 两个重要的图像分析工具：直方图、散点图 3.遥感数字图像增强处理 图像增强：概念、方法 空间域增强、频率域增强 3.1辐射增强：概念、实现原理 直方图修正，线性变换、分段线性变换算法原理 直方图均衡化、直方图匹配的应用 3.2空间增强 邻域、邻域运算、模板、模板运算 空间增强的概念 平滑（均值滤波、中值滤波）原理、特点、应用 锐化、边缘增强概念
方向模板、罗伯特算子、索伯尔算子、拉普拉斯算子的算法和特点? 计算图像经过下列操作后，其中心象元的值： – 3×3中值滤波 –采用3×3平滑图像的减平滑边缘增强 –域值为2的3×1平滑模板 – Sobel边缘检测 – Roberts边缘检测 –模板 3.3频率域处理 高频和低频的意义 图像的傅里叶频谱 频率域增强的一般过程 频率域低通滤波 频率域高通滤波 同态滤波的应用 3.4彩色增强 彩色影像的类型：真彩色、假彩色、伪彩色
江苏省2017年普通高校专转本选拔考试 计算机基础试题卷 注意事项： 1本试卷分为试题卷和答题卡两部分，试题卷共 8 页。全卷满分 100 分，考试时间 90 分钟。 2.必须在答题卡上作答，作答在试题卷上无效。作答前务必将自己的姓名和准考证号准确清晰地填写在试题卷和答题卡上的指定位置。 3.考试结束时，须将试题卷和答题卡一并交回。 一、判断题（本大题共 10 小题，每小题 1 分，共计 10 分。下列每小题表述正确的在答题卡上将 A 涂黑，错误的将 B 涂黑） 1.2017 年春节期间，QQ 推出了一款新的抢红包功能，用户可以利用手机的摄像头，在任何生活环境中去寻找红包。该新功能主要采用了虚拟宣誓技术。( ) 2.截止到2016 年12 月，全球超级计算机运算速度排名第一的是我国自主研制的天河二号。( ) 3.内存中的数据是按地址存取的。( ) 4.对U 盘的格式化，会清楚用户存储在该盘上的所有数据。( ) 5.算法和程序都必须在执行了有穷步的操作后终止。( ) 6.开放系统互连参考模型（OSI/RM）将网络分为网络接口层、网络层、运输层和应用层。( ) 7.Internet 是当前全球最大的、开放的、由众多网络相互连接而成的特定计算机网络，它采用TCP/IP 协议族作为通信的规则。( ) 8.汉字在计算机中可以用宋体、楷体等不同字体显示。在GB18030 汉字编码中，宋体字和楷体的“平”字对应的机内码不相同。( ) 9.图像数字化后每个像素所占二进制位数就是该图像所包含的颜色数。( ) 10.在数据库设计时，常用E-R 图对用户的需求进行抽象表示，E-R 图与具体的系统及软硬件环境无关。( ) 二、选择题（本大题共50 小题，每小题1 分，共计50 分。在下列每小题中，选出一个正确答案，并在答题卡上将所选项的字母标号涂黑） 11.用十六进制形式表示8 位二进制整数时，需要的位数是( ) A.1 B.2 C.3 D.4 12.两个无符号二进制数进行减法运算，—00001001 的结果是( ) A.00001000 B. C. D. 13.下列不同进制的无符号数中，数值最大的是( ) A. B.167D C.251Q D.ADH 14.若在一个非零无符号二进制整数右侧添加3 个0，则新数值是原数值的( ) A.1/8 倍 B.1/3 倍 C.3 倍 D.8 倍 15.将某8 位二进制的低 4 位清零，高4 位保持不变，下列操作正确的是( )
《计算机图形图像处理》教学基本要求 适用专业：计算机动画设计（高技） 课程类别：专业主干课（项目课程）+专业实践课 参考学时：246学时（包含专业实践210学时） 学分：2+12=14 参考教材：《平面设计Photoshop 7.0》. 王维编著. 华东师范大学出版社教学参考：《Photoshop图像处理项目教程》.陆一琳主编. 华中科技大学出版社编写执笔人：肖进审定负责人：徐海 一、课程性质和任务 1、课程性质： Photoshop 是Adobe 公司推出的一款目前非常流行、应用非常广泛的图片处理软件。伴随着计算机的普及和计算机在各行业的广泛应用，Photoshop 发挥了越来越大的作用。计算机和数码相机的普及，使用者可以在家中进行简单的图片处理，这使得Photoshop 可以作为一个应用软件在所有学生中推广。社会上各种数码冲印、数码影楼、数码海报广告的出现也直接为很好学习Photoshop 的学生提供了就业机会。 Photoshop 作为图片处理软件，现在可以作为所有非计算机专业学生的选修课；同时Photoshop 具备非常强大的图片处理功能，能很好地为动画、多媒体、网页制作等等提供经过处理制作的图片素材，图片处理的好坏直接关系到作品的美观效果，是计算机专业的学生必修的一门课程。 2、课程任务： Photoshop 教学过程中应注重培养学生的思考和动手能力，把知识点穿插在实例中进行教学，一方面启迪学生去思考实例是如何实现的，另一方面让学生通过操作完成实例的创作。使学生在轻松愉快的过程中完成学习任务，掌握Photoshop 的使用。教师应重视实例的选择，要求实例能突出新知识点，同时也兼顾旧知识点，操作的难度要适中，通过教学过程中的启迪和帮助能够完成教
(单选题)1: 下列哪种格式用于网页中的图像制作：_______ A: EPS B: DCS 2.0 C: TIFF D: JPEG 正确答案: D (单选题)2: 当使用JPEG 作为优化图像的格式时：________ A: JPEG 虽然不能支持动画，但它比其它的优化文件格式（GIF和PNG）所产生的文件一定小B: 当图像颜色数量限制在256 色以下时，JPEG 文件总比GIF 的大一些 C: 图像质量百分比值越高，文件越大 D: 图像质量百分比值越高，文件越小 正确答案: C (单选题)3: Alpha 通道相当于几位的灰度图：_______ A: 4位 B: 8位 C: 16位 D: 32位 正确答案: C (单选题)4: 下列可以使图像产生立体光照效果的滤镜是_______ A: 风 B: 等高线 C: 浮雕效果 D: 照亮边缘 正确答案: C (单选题)5: 下列哪四种模式可使用16位/通道来代替默认的8位/通道？ A: 灰度、RGB、CMYK或Lab模式 B: 位图、灰度、CMYK或Lab模式 C: 灰度、双色调、CMYK或Lab模式 D: 灰度、索引颜色、RGB或Lab模式 正确答案: A (单选题)6: 下列选项哪些是动作调板与历史记录调板都具有的的特点______ A: 在关闭图像后所有记录仍然会保留下来 B: 都可以对文件夹中的所有图像进行批处理 C: 虽然记录的方式不同，但都可以记录对图象所做的操作 D: 历史调板记录的信息要比动作调板广 正确答案: C (单选题)7: 移动图层中的图像时，如果每次需移动10 个象素的距离，应______ A: 按住Alt键的同时按键盘上的箭头键
Photoshop在计算机图形图像处理中的应用 发表时间：2017-07-27T15:29:09.987Z 来源：《基层建设》2017年第10期作者：黄艳华 [导读] 摘要：随着科学技术的飞速发展，对计算机图像处理技术的使用已经渐渐地趋向成熟，计算机图像处理被广泛的应用到多个领域廊坊市高级技工学校河北省廊坊市 065000 摘要：随着科学技术的飞速发展，对计算机图像处理技术的使用已经渐渐地趋向成熟，计算机图像处理被广泛的应用到多个领域，推动了社会的进步和发展。计算机图形处理技术在工业、农业、广告传媒等领域被广泛的应用。本研究中主要分析计算机图形图像处理中 Photoshop的应用。 关键词：Photoshop；计算机图形图像；处理应用 引言 由于信息技术的发展，计算机技术得到了广泛的应用，社会的生产方式和人们的生活方式都发生了巨大的变化，计算机图书处理技术的功能较高，有着重要的社会价值。随着社会的发展，人们的经济水平和文化素质都得到了相应的提高，对图像的要求也随之提高。计算机图形处理技术要不断地提高技术水平，创造出更加清晰的图像来满足人们的需求。 1、Photoshop软件应用 计算机图形图像处理技术主要分为：三维制作技术方面图形处理和二维制作技术方面图像处理。在软件制作上也主要从这两个方向进行开发，西方国家的图形图像处理技术处于领先地位，后期在全球广泛应用开发过程中，这项技术得到推广。图形处理在三维制作技术的基础上在电脑上制作各种物体，添加一系列动作，通过整体优化完成完整的图形处理。 图像处理的应用更为普遍，在已有的图像基础上进行变化时，通过不同功能要求添加效果，这种图像处理技术在日常应用中简称PS技术，常见的拍照、图片处理都会有相类似的软件处理应用，这项技术的推广和普及在现代审美要求上满足了人们的需求。Photoshop软件的处理技术在图形图像处理上更为专业化，从软件开发历史来看，计算机技术的不足为这项技术软件的开发提出了问题，托马斯?诺尔和约翰?诺尔兄弟在计算机黑白图像的显示问题上编写的程序Display在不断修改的基础上成为更为强大的图像编辑程序，经过更名成为现在的Photoshop；在结合商用仪器的推广中，与Adobe公司的合作成为跳板，得到更为广泛的应用推广。 2、Photoshop实际运用分析 2.1 套索工具 套索工具的作用就是在拼接各项图片素材时,想要对特定的区域进行处理,就需要使用套索工具选取需要编辑的部位,这也就是俗称的选取功能。Photoshop中的套索工具分为三种,套索工具、多边形套索工具、磁性套索工具。这三种不同的套索工具适用于使用者对图片的不同要求。套索工具就类似于画笔,使用者只需要移动鼠标将需要选择的区域圈起来即可,这种方式适用于选择边界较为复杂且不是直线的图片。而多边形套索工具可以通过点击鼠标左键来定位,每两个定位之间由软件自主生成一条选取线,围绕一圈后完成选区,这种方式主要适用于图片中边线较直且鼠标难以保持直线的图形选择。磁性套索工具是Photoshop上一种较为新型的套索工具,这种工具可以使用鼠标规划大致区域,软件可以自主对图像进行识别,向选取边界靠拢,这种方式使用较多,而且能够满足大部分图片的需要,选取精度较高。 2.2 光度色彩曲线调整 一张图片中存在一定的光度和色彩,在Photoshop中,可以通过调整曲线命令来进行光度和色彩调整。通常来说,在某些照片的拍摄过程中,由于各种因素的影响,容易导致照片出现曝光度过高、色彩反差太小等情况,这时候可以将光度曲线调低,将图片的光亮值进行重新分布,将需要提亮的位置光度调高,而需要暗淡的位置可以适当降低光度曲线,从而使图片有更加适宜的光亮度。 2.3 线性减淡 众所周知,在进行图片处理时,一些图片有一定的背景痕迹干扰,尤其是在进行图片扫描时,各种皱褶、指纹、笔记、图表、logo等图案,都会对我们的图片处理工作造成影响。就Photoshop中,可以利用线性减淡技术将土层中的干扰图案进行减淡、甚至消除,这种方式比现有的照相处理效果要更好。在图片的RGB通道中,所有的黑色字体或图案以及干扰项都占有同一个比例,在操作时就可以选择RGB通道中干扰项所占据的位置,对其进行线性减淡,从而达到消除干扰的目的。 3、Photoshop在计算机图形图像处理中的应用技巧 Photoshop作为在全球图形图像处理软件中使用率最高的软件，在发展中把数字化摄影图片、剪辑、绘画、图形及现有的美术作品结合在一起，在熟悉各种功能的过程中培养艺术审美能力对图像处理技术的提升是有利的。在Photoshop的多种技术中，通道技术、蒙版技术、滤镜技术等都是比较常用的基本操作。在不同的工具操作和应用下，对图像处理的效果也各不相同。 在Photoshop软件图像处理上有一些使用技巧对更高效地完成图像处理工作是有帮助的，在分辨率、图像裁切和缩放、利用通道工具创建图像特效等技巧上都有需要注意的细节问题。分辨率对图像来说是一项重要指标，在Photoshop软件中对图像画面质量处理上的应用有很大不同，同一张图像，在分辨率上的差异通过数字可体现出来，分辨率的高低决定着像素的大小和图片的质量问题，因此在图像处理上合理选择分辨率对图像质量和效果有重要影响。同时图像的裁切和缩放在Photoshop软件技术中是一项重要方法，调整画布大小，按照标尺选择想要裁剪的像素，完成裁剪。缩放图像在图像处理中也是常用的方法。Photoshop的通道功能在PS中是一项重要功能，图像的特效处理都是利用通道技术来完成的，根据颜色信息储存不同分为：颜色通道、Alpha通道、专色通道。此外，在曲线调整上，Photoshop中有很多可以调整图片亮度及颜色方法，不同曲线调整产生不同亮度效果在图像处理技术中作用明显。图像处理对污点的处理研究也在不断完善，Photoshop软件可以消除图像上的污点，利用软件中的相关修复功能，自动将修复区域进行整体效果恢复。 Photoshop在使用技巧上需要更多的操作实践和耐心进行研究，在图像处理上同样需要有技术上和技巧上的潜心研究和不断探索。Photoshop在计算机技术的飞速发展中对时代应用需求的不断把握，在功能和运营上都会及时作出调整，未来在图像处理上也会有更多的创意融入到软件的完善中。 4、结语 Photoshop是一项强大的处理软件,本文简要介绍了Photoshop的一些基本功能,并结合动画设计中所需要的功能进行实用性探讨,但由于篇幅有限,只能大致阐述其功能,但是Photoshop的功能远不止于此,其中还包含了许多更加强大的功能。正是由于它无可比拟的功能性和实用性,不仅在设计行业深受青睐,甚至在整个计算机行业都受到了广泛关注。随着科技的不断发展,Photoshop的应用还有更加广阔的发展空间等
《数字图像处理》课程设计 课设题目：图像增强与MATLAB实现学校学院：华东交通大学理学院 学生班级：13级信息计算（2）班学生：超 学生学号：20130810010216 指导老师：自柱
图像增强与MATLAB实现 摘要 数字图像处理是指将图像信号转换成数字格式并利用计算机对其进行处理的过程。图像增强是数字图像处理的过程中经常采用的一种方法，它对提高图像质量起着重要的作用。本文先对图像增强的原理进行概述，然后对图像增强的方法分类并给出直方图增强、对比度增强、平滑和锐化等几种常用的增强方法的理论基础，通过Matlab实验得出的实际处理效果来对比各种算法的优缺点，讨论不同的增强算法的技术要点，并对其图像增强方法进行性能评价。 关键字：图像；图像增强；算法
目录 一、MATLAB的简介 (1) 1.1MATLAB主要功能 (1) 二、MATLAB的主要功能 (1) 2.1数字增强技术概述 (1) 2.2数字图像的表示 (2)
三、直方图的均衡化 (2) 3.1图像的灰度 (2) 3.2灰度直方图 (2) 3.3直方图均衡化 (3) 四、图像二值化 (5) 4.1图像二值化 (5) 五、对比度增强 (7) 5.1对比度增强 (7) 5.2灰度调整 (8) 5.3对数变换 (9) 六、滤波 (10) 6.1平滑滤波 (10) 6.2线性平滑滤波程序： (11) 6.3非线性滤波 (12) 七、锐化 (18) 八、参考文献 (19) 九、自我评价 (20)
一、Matlab的简介 1.1 MATLAB主要功能 MATLAB是建立在向量、数组和矩阵基础上的一种分析和仿真工具软件包，包含各种能够进行常规运算的“工具箱”，如常用的矩阵代数运算、数组运算、方程求根、优化计算及函数求导积分符号运算等；同时还提供了编程计算的编程特性，通过编程可以解决一些复杂的工程问题；也可绘制二维、三维图形，输出结果可视化。目前，已成为工程领域中较常用的软件工具包之一。 二、MATLAB的主要功能 2.1数字增强技术概述 图像增强是按特定的需要突出一幅图像中的某些信息，同时，消弱或去除某些信息使得图像更加实用。图像增强技术主要包含直方图修改处理、图像平滑处理、图像尖锐化处理等。 图像增强技术主要包括：直方图修改处理，图像平滑处理，图像尖锐化处理，彩色图像处理。从纯技术上讲主要有两类：频域处理法和空域处理法。 频域处理法主要是卷积定理，采用修改图像傅立叶变换的方法实现对图像的增强处理技术；空域处理法：是直接对图像中的像素进行处理，基本上是以灰度映射变换为基础的。
实验报告 班级：08108班 姓名：王胤鑫 09号 学号：08210224 一、实验内容 给出噪声图像，请选择合适的图像增强算法，给出你认为最优的增强后的图像。 可以使用Matlab - Image Processing Toolbox 中的处理函数。 原始图像如下： 二、算法分析 对于给出的图像中有灰色的噪声，因此首先处理灰色的线条，根据其方差的大小来判断其所在行。对于两条白色的噪声，根据与前后两行的对比来判断其所在位置。程序中设定灰色线条处理的均方差门限为，白线处理的标准为与前后两行的差值超过（转换为double型）。滤除噪声之后再通过中值滤波、拉普拉斯图像增强等方式对图像进行处理。
三、matlab 源程序 clear all;clc; f=imread(''); figure,imshow(f),title('原始图像'); [m,n]=size(f); f0= im2double(f); % 整型转换为 double 类 f1=f0; std_i=zeros(1,m-2); %灰线处理 for i=2:m-1 %灰线处理 std_i(i-1)=std(f0(i,:)); if(std_i(i-1)< for j=1:m f0(i,j)=(f0(i-1,j)+f0(i+1,j))/2; end end end figure,imshow(f0),title('滤除灰线后的图像'); fz=f0-f1; [r,c]=find(fz~=0);%寻找灰线噪声的位置 f2=f0; change=0; count=0; for i=3:m-2 %白线处理 for j=1:m if(abs(f0(i,j)-f0(i-1,j))>&&abs(f0(i,j)-f0(i+1,j))> count=count+1; end if(count>n* count=0; change=1; break; end end if(change==1) for k=1:m f0(i,k)=(f0(i-1,k)+f0(i+1,k))/2; end change=0; count=0; end end
计算机图形图像处理试题(A)(含答案)
计算机图形图像处理试题(A) 班级_____________姓名____________ 学号_________________ 一、选择(每小题2分，共30分) 1、下列哪个文件不属于位图图形文件格式________。 A、PSD B、TIFF C、DWG D、BMP 2._______图层上的图形虽然可以显示，但却不能被修改。 A、关闭 B、冻结 C、锁定 D、透明 3.在对图形进行旋转时，下列说法正确的是______。 A、角度为正时，将逆时针旋转 B、角度为负时，将逆时针旋转 C、角度为正时，将顺时针旋转 D、旋转方向不受角度正负值影响 4、Photoshop中绘制圆形选区时，先选择椭圆选框工具，在按下______的同时，拖动鼠标，就可以实现圆形选区的创建。 A、Alt键 B、Ctrl键 C、Shift键 D、Ctrl+Alt 5、下列哪个内部滤镜可以实现立体化效果_______。
A、风 B 等高线 C 浮雕效果 D 撕边 6、哪种方法不能打开一个图形文件_______。 A、Ctrl+O B、双击工作区域 C、直接从外部拖动一幅图片到Photoshop界面上 D、Ctrl+n 7、Photoshop中设置一个适当的羽化值，然后对选区内的图形进行多次Del操作可以实现_______。 A、选区边缘的锐化效果 B、选区边缘的模糊效果 C、选区扩边 D、选区扩大 8、Photoshop中在默认情况下使用缩放工具时，按住Alt键的同时点击鼠标左键，可以实现_______。 A、图像像素匹配 B、图像大小适合屏幕 C、图像缩小 D、图像放大 9、Photoshop中要同时移动多个图层，则需先对它们进行_______操作。 A、图层链接 B、图层格式化 C、图层属性设置 D、图层锁定 10、下列哪个不属于在图层面板中可以调节的参数_______。 A、透明度 B、编辑锁定 C、显示隐藏当前图层 D、图层的大小 11、为了确定磁性套索工具对图像边缘的敏感程度，应调整下列哪个数值 ______。 A、容差 B、边缘对比度 C、颜色容差 D、套索宽度 12.如果给定圆心和起点，，下列指定的________不能画出精确的圆弧。 A、端点 B、角度 C、长度 D、方向
第一单元选区 1.1 第1题 【操作要求】 制作立体饼状物，如图X1-1所示。 图X1-1 效果图 1.建立选区：建立直径为230像素的圆形，制作饼状比例示意图选区如图Y1-1所示。 2.选区编辑：填充270°绿色（#0a9204）、50°红色（#a40603）和40°蓝色（#1709dc）。 3.效果装饰：制作立体效果（30像素高度）。 将最终结果以Xl-l.psd为文件名保存在考生文件夹中。 图Y 1-1
【操作要求】 合成窗外风景效果，如图X1-2所示。 图X1-2 效果图 1.建立选区：打开C:\2007PS\Unit1\window.jpg和green.jpg素材文件，如图所示。 2.选区编辑：把风景图像合成到窗口。 3.效果装饰：调整透过窗口看到两棵大树的位置。 将最终结果以X1-2.psd为文件名保存在考生文件夹中。
【操作要求】 制作连体字，如图X1-3所示。 图1-3 效果图 1.建立选区：输入130像素、宋体、斜体字符“2（#340fea）、0（#74ebef)、0（#86e32e）、0（#ecec00）”并做立体化处理，如图Y1-3所示。 2．选区编辑：制作相互嵌套的连环字符。 3.效果装饰：添加红色（#d7042f）前景。 将最终结果以X1-3.psd为文件名保存在考生文件夹中。 图Y1-3
【操作要求】 制作三色谱，如图X1-4所示。 图X1-4效果图 1.建立选区：制作直径为85像素的红色（#f60400）、绿色（#08fb01）和蓝色（#0100fe）三个圆形，如图Y1-4所示。 2.选区编辑：将红蓝圆形选区交集填充为洋红色（f7004fe）、将红绿圆形相交处填充为黄色（#feff01）、将绿蓝圆形相交处填充为青色（#09fbff），将红绿蓝三圆相交处填充为白色（#ffffff）。 3.效果装饰：调整圆形交集大小等比，并且调整圆形至圆心处位置。 将最终结果以X1-4. psd为文件名保存在考生文件夹中。 图Y1-4
实验六遥感影像增强处理 实习目的：掌握常用的遥感影像增强处理的方法。 实习内容：遥感影像空间、辐射、光谱增强处理的主要方法 空间增强：包括卷积增强处理、纹理分析、自适应滤波等 辐射增强：LUT拉伸处理、直方图均衡化处理、直方图匹配、亮度反转处理等 光谱增强：主成份变换、缨帽变换、色彩变换、指数计算等 图像增强是改善图像质量、增加图像信息量、加强图像判读和识别效果的图像处理方法。图像增强的目的是针对给定图像的不同应用，强调图像的整体或局部特性，将原来不清晰的图像变得清晰或增强某些感兴趣区域的特征，扩大图像中不同物体特征之间的差别，满足某些特殊分析的需要。图像增强的途径是通过一定的手段对原图像附加一些信息或变换数据，有选择的突出图像中感兴趣区域的特征或抑制图像中某些不需要的特征。图像增强的方法包括空间域增强和频率域增强两类。空间域增强包括空间增强、辐射增强和光谱增强。在实际运用中，不是所有的图象增强处理方法都要用到，具体采用哪种图象增强处理方法，视具体的研究区域，研究内容和对象而定。 1．图像解译功能简介（Introduction of Image Interpreter） 利用ERADS IMAGINE 进行图像增强主要采用ERADS IMAGINE的图像解译器（Image Interpreter）模块，该模块包含了50多个用于遥感图像处理的功能模块，这些功能模块在执行过程中都需要通过各种按键或对话框定义参数，多数功能都借助模型生成器（Model Maker）建立了图形模型算法，容易调用或编辑。 图像解译器(Image Interpreter或Interpreter)，可以通过两种途径启动：ERDAS图标面板菜单条： Main/Image Interpreter--Image Interpreter 菜单 ERDAS图标面板工具条：点击Interpreter图标一Image Interpreter菜单
课程名称： 实验项目： 实验地点： 专业班级：学号：学生姓名： 指导教师： 2012年月日
实验一 空域图像增强技术 一、 实验目的 1结合实例学习如何在视频显示程序中增加图像处理算法； 2理解和掌握图像的线性变换和直方图均衡化的原理和应用； 3了解平滑处理的算法和用途，学习使用均值滤波、中值滤波和拉普拉斯锐化进行图像增强处理的程序设计方法； 4 了解噪声模型及对图像添加噪声的基本方法。 二、 实验原理 1 灰度线性变换就是将图像中所有点的灰度按照线性灰度变换函数进行变换。 )],([),(y x f T y x g = ?? ???<≤+-<≤+-≤≤=255),(]),([),( ]),([),(0 ),(),(y x f b g b y x f b y x f a g a y x f a y x f y x f y x g b a γβα n y m x ,2,1 ,,,2,1== 2 直方图均衡化通过点运算将输入图像转换为在每一级上都有相等像素点数的输出图像。按照图像概率密度函数PDF 的定义： 1,...,2,1,0 )(-==L k n n r p k k r 通过转换公式获得： 1,...,2,1,0 )()(00-====∑∑==L k n n r p r T s k j k j j j r k k 3 均值（中值）滤波是指在图像上，对待处理的像素给定一个模板，该模板包括了其周围的临近像素。将模板中的全体像素的均值（中值）来代替原来像素值的方法。 4 拉普拉斯算子如下： ???? ??????--------111181111 拉普拉斯算子首先将自身与周围的8个像素相减，表示自身与周围像素的差异，再将这个差异加上自身作为新像素的灰度。 三、 实验步骤 1 启动MATLAB 程序，对图像文件分别进行灰度线性变换（参考教材57页，例4.1）、直方图均衡化、均值滤波、中值滤波和梯度锐化操作。添加噪声，重复上述过程观察处理结果。 2记录和整理实验报告
计算机图形图像处理试题(B) 班级_____________姓名____________ 学号_________________ 一、选择(每小题2分，共30分) 1、关于RGB正确的描述的是________. A、色光三元色 B、印刷用色 C、一种专色 D、网页用色 2、假设坐标点的当前位置是(300,200)，现在从键盘上输入了新的坐标值(@-100,200)，则新的坐标位置是________。 A、(200,400) B、(-100,200) C、(300,200) D、(100,200) 3、使用“直线”LINE命令绘图，在连续输入了三个点之后，再输入字符______将绘制出封闭的图形。 A、A B、U (Undo) C、C (Close) D、E 4、Photoshop中我们通常使用_______工具来绘制路径。 A、钢笔 B、画笔 C、喷枪 D、选框 5、绘制标准五角星使用________工具。 A、喷枪 B、多边形索套 C、铅笔 D、多边形 6、给选择区添加前景色使用_______组合键。 A、CTRL+DEL B、ATL+DEL C、SHIFT+DEL D、回车 7、下面的描术述中，______不符合Photoshop图层的描述。 A、用鼠标单击一个图层，可以使之高这度显示 B、用鼠标双击一个图层，可以打开图层选项面板 C、把图层直接拖到新图层按钮上可以复制图层 D、图层上的物件不可以随便修改 8、在PHOTOSHOP 的PSD格式描述正确的是_______。 A、是PHOTOSHOP自己的格式 B、可以通用的格式 C、一种图形格式 D、可以网络使用的格式 9、PHOTOSHOP 工具箱的工具中有黑色向右的小三角符号，表示_______。 A、表示可以选出菜单 B、表示能点出对话框 C、有并列的工具 D、表示该工具有特殊作用 10、可以还原多步操作的面板是_______。 A、action 面板 B、路径面径 C、历史记录面板 D、图层面板 11、在CAD中创建一个圆与已知圆同心，可以使用哪个修改命令________。 A、阵列 B、复制 C、偏移 D、镜相 12、在CAD中要始终保持物体的颜色与图层的颜色一致，物体的颜色应设置为_______。 A、按图层 B、图层锁定 C、按颜色 D、按红色 13、在CAD中SA VE命令可以________。 A、保存图形 B、会退出AutoCAD C、定期地将信息保存在磁盘上 D、以上都是 14、在CAD中如果可见的栅格间距设置得太小，AutoCAD将出现如下提示________。
实验三：遥感图像的增强处理 (3机时) 实验目的：通过上机操作，了解空间增强、辐射增强几种遥感图象增强处理的过程和方法，加深对图象增强处理的理解。 实验内容：卷积增强处理；锐化增强处理；直方图均衡化；色彩变换。 ERDAS IMAGE图像解译模块主要包括了图像的空间增强、辐射增强、光谱增强、高光谱工具、傅立叶变换、地形分析以及其他实用功能。 实验数据：wx98tm543.img（待校正图像）与wx98spot_pan.img（参考图像）校正的结果 wx98tm543_warp.img；ERDAS安装目录中的若干样例图像数据文件。 1、卷积增强（Convolution） 空间增强技术是利用像元自身及其周围像元的灰度值进行运算，达到增强整个图像之目的。卷积增强（Convolution）是空间增强的一种方法。 卷积增强（Convolution）时将整个像元分块进行平均处理，用于改变图像的空间频率特征。卷积增强（Convolution）处理的关键是卷计算子----系数矩阵的选择。该系数矩阵又称卷积核（Kernal）。ERDAS IMAGINE将常用的卷计算子放在一个名为default.klb的文件中，分为3*3，5*5、7*7三组，每组又包括“EdgeDetect/Low Pass/Horizontal/Vertical/Summary”等七种不同的处理方式。具体执行过程如下： ERDAS图标面板菜单条：Main→Image Interpreter→Spatial enhancement →convolution→convolution对话框。
图3-1 Convolution对话框 几个重要参数的设置： 边缘处理方法：（Handle Edges by）：Reflection 卷积归一化处理：Normalize the Kernel 2、直方图均衡化（Histogram Equalization） 直方图均衡化实质上是对图像进行非线性拉伸，重新分配图像像元值，是一定灰度范围内的像元数量大致相同。这样，原来直方图中间的峰顶部分对比度得到增强，而两侧的谷底部分对比度降低，输出图像的直方图是一较平的分段直方图。注意：认真对比直方图均衡化前后图像差别，仔细观察直方图均衡化的效果。 图3-2直方图均衡化 3、主成分变换 主成分变换（Principal Component Analysis）是一种常用的数据压缩方法，它可以将具有相关性的多波段数据压缩到完全独立的较少的几个波段上，使图像数据更易于解译。ERDAS IMAGE提供的主成分变换功能最多等对256个波段的图象进行转换压缩。 ERDAS 图标面板菜单条：Main →Image Interporeter→Spectral Enhancement →Principial Comp →Pincipal Components对话框。（图3-3）
计算机图像处理学习周期 02任务-在线作业答案 100分 2010年06月10日星期四下午 5:09 一、单项选择题（共 15 道试题，共 90 分。） 1. 打开文字输入工具的快捷方式是按下键盘上的：（ A ） A. T B. X C. Z D. V 2. 下面对文字格式设置功能叙述有错误的是：( D ) A. 可设置文字间距 B. 可用来设置文字缩放 C. 可设置文字的形状变化 D. 不能用来设置文字的上、下标 3. 所有的滤镜都不可使用（设图像是8位/通道）的图像模式是：( B ) A. CMYK B. 索引颜色 C. 多通道 D. 灰度 4. 下列关于滤镜的操作原则正确的是：（ B ） A. 滤镜不仅可用于当前可视图层，对隐藏的图层也有效。 B. 不能将滤镜应用于位图模式（Bitmap）或索引颜色（Index Color）的图像。 C. 滤镜只对RGB图像起作用。 D. 滤镜不可用于16位/通道图像。 5. 模糊滤镜包括：（ A ） A. 动感模糊滤镜、径向模糊滤镜、模糊滤镜、特殊模糊滤镜、进一步模糊滤镜、高斯模糊滤镜 B. 动感模糊滤镜、特殊模糊滤镜、进一步模糊滤镜、高斯模糊滤镜 C. 动感模糊滤镜、径向模糊滤镜、模糊滤镜、特殊模糊滤镜 D. 动感模糊滤镜、径向模糊滤镜、模糊滤镜、模糊滤镜、高斯模糊滤镜 6. 下面是关于“抽出”滤镜的叙述，正确的是：（ B ） A. “抽出”滤镜用于选取对象 B. “抽出”滤镜用于在图像中分离对象 C. “抽出”滤镜的操作就是描边 D. “抽出”滤镜的操作就是填充 7. 下面是关于“像素化”滤镜的叙述，不正确的是：（ D ） A. “像素化”滤镜处理后的图像是由颜色块构成的 B. “马赛克”滤镜属于像素化”滤镜，它可以设置马赛克的大小 C. “铜版雕刻”滤镜适宜模拟不光滑或年代已久的金属效果
实验二图像增强处理实习报告 1.实验目的和内容 1.1.实验目的 掌握图像合成和显示增强的基本方法，理解存储的图像数据与显示的图像数据乊间的1.2.实验要求 熟练根据图像中的地物特征进行图像合成显示、拉伸、图像均衡化等显示增强操作。 理解直方图的含义，能熟练的利用直方图进行多波段的图像显示拉伸增强处理。 1.3.软件和数据 ENVI 软件。 TM 图像数据。上次实验合成后的图像数据文件AA。 1.4.实验内容 图像的彩色合成显示 图像的基本拉伸方法 图像均衡化方法 图像规定化 2.实验过程 通过合成和拉伸增强显示图像中的信息。 2.1.图像合成 图像合成方法：伪彩色合成、彩色合成两种方式。其中彩色合成包括：真彩色合成、假彩色合成、模拟真彩色合成。 操作： 使用（4，3，2）进行RGB 合成显示图像。图像窗口为#1。
移动图像窗口的红色选框到玄武湖，将光标十字放在红框内，双击，显示光标位置窗口。该窗口中出现了Scrn 和Data，二者后面的RGB 的值是不同的。
2.1.1伪彩色合成 在新的窗口显示第4 波段图像，窗口为#2。
操作： 菜单：窗口菜单Tools-Color Mapping-Density slice…，选择Band 4，确定。在“Density Slice”窗口中，点击“应用”按钮，窗口#2 的图像变成了彩色。
设置默认的分级数为3 个：在“Density Slice”窗口，点击Options-Set number of default range,输入3，确定。点击Options-Apply default range，点击Apply 按钮。查看窗口#2 内的变化。
一、实验名称 遥感图像光谱增强处理 二、实验目的 对图像进行主成分分析、主成分变换以及主成分百分比计算；观察图像在不同色彩空间之间相互转换的结果异同，对图像进行融合，用MODEL MAKER 建模方式进行图像处理。 通过以上操作初步掌握图像光谱增强处理过程，进一步理解影像光谱增强中不同增强方法的原理及其增强效果的差异。 三、实验原理 光谱增强是基于多光谱数据对波段进行变换达到图像增强处理，采用一系列技术去改善图象的视觉效果，或将图象转换成一种更适合于人或机器进行分析处理的形式。有选择地突出某些对人或机器分析有意义的信息，抑制无用信息，提高图象的使用价值。 主成分分析（PCA）用多波段数据的一个线性变换，变换数据到一个新的坐标系统，以使数据的差异达到最大。对于增强信息含量、隔离噪声、减少数据维数非常有用。 使用Color Transforms 工具可以将3-波段红、绿、蓝图像变换到一个特定的彩色空间，并且能从所选彩色空间变换回RGB。两次变换之间，通过对比度拉伸，可以生成一个色彩增强的彩色合成图像。 图像融合是将多幅影像组合到单一合成影像的处理过程。它一般使用高空间分辨率的全色影像或单一波段的雷达影像来增强多光谱影像的空间分辨率。 四、数据来源 本次实验所用数据来自于国际数据服务平台；landsat4-5波段30米分辨率TM第三波段影像，投影为WGS-84，影像主要为山西省大同市恒山地区，中心纬度：38.90407 中心经度：113.11840。
五、实验过程 1.主成分分析 1）打开并显示TM影像文件，从ENVI 主菜单中，选择File →Open Image File选择影像，点击Load Band 在主窗口加载影像。 2）主菜单选择Transforms—>Principal Components—>Forward PC Rotation —>Compute New Statistics and Rotate。在弹出的Principal Components Input File 对话框中，选择图像。 3)在Forward PC Rotation Parameters对话框中在输入统计系数，选择计算矩阵（选择协方差矩阵），输出统计文件及路线，统计波段数等相关参数的设置，单击Ok。