CN104754340B - Unmanned aerial vehicle reconnaissance image compression method - Google Patents
Unmanned aerial vehicle reconnaissance image compression method Download PDFInfo
- Publication number
- CN104754340B CN104754340B CN201510102937.4A CN201510102937A CN104754340B CN 104754340 B CN104754340 B CN 104754340B CN 201510102937 A CN201510102937 A CN 201510102937A CN 104754340 B CN104754340 B CN 104754340B
- Authority
- CN
- China
- Prior art keywords
- image
- aerial vehicle
- unmanned aerial
- region
- reconnaissance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Remote Sensing (AREA)
- Image Processing (AREA)
Abstract
The invention discloses an unmanned aerial vehicle scout image compression method, which is characterized in that most of objects targeted by an unmanned aerial vehicle are military vehicles and pillboxes which are objects with regular shapes, wherein before compression, an interested scout target area and a background area of an unmanned aerial vehicle scout image are completely separated without mutual influence, and suitable compression methods can be freely selected for the two parts. The unmanned aerial vehicle scout image compression method separately selects different bit rates for the region of interest and the background region, and ensures that the region of interest scout target is compressed by a small compression ratio and the background region is compressed by a large compression ratio. Meanwhile, the segmentation compression method can also freely select the size of the compression ratio, and the image quality of the target and the background can be freely adjusted. And the compression bit rate of the reconnaissance target and the background area can be freely adjusted according to the requirement of the transmission bandwidth of the equipment, so that the compression requirement of the reconnaissance image of the unmanned aerial vehicle is met. Meanwhile, the better the recovery quality of the region is, the larger the peak signal-to-noise ratio of the corresponding region is.
Description
The technical field is as follows:
the invention relates to an unmanned aerial vehicle reconnaissance image compression method, and belongs to the field of unmanned aerial vehicle reconnaissance.
Background art:
the reason why the image can be compressed is that the data information of the image contains a large amount of redundant information in addition to the effective information. In practical applications, people often pay attention to only a certain part of the contents, but are not interested in the contents of the rest parts. That is, when processing an image, the degree of interest corresponding to different regions of the entire image is different depending on the purpose of the user's subjective application. However, obtaining useful information from images still requires a lot of information on non-interesting areas. Then, when encoding, neither the same encoding method can be applied to the whole image, nor only the interested content can be transmitted. Therefore, the optimal coding effect can be obtained only by performing different coding processes on the interested region and the non-interested region of the image. In practice, people always want the content concerned by themselves to be compressed better than the rest.
Region of interest (ROI) -based image compression coding is a research hotspot and difficulty in the field of current compression coding. The method can realize that the compression quality of the image interesting region is higher than that of the image background region, and can meet the requirements of people on the image. Research based on region-of-interest coding techniques has achieved many valuable research results, especially the region-of-interest (ROI) coding standard proposed in the JPEG2000 standard, namely, the Max-shift method (Max-shift) and the general-shifting-based (Generic-scaling-based), which lays a solid foundation for the development of region-of-interest-based coding techniques.
Much foreign research on region-of-interest compression mainly focuses on the improvement of wavelet coefficient translation idea and the combination of two methods in the standard and different coding algorithms. Domestic research on region-of-interest coding is also increasing, but most of the research focuses on the JPEG2000 standard and is still in the preliminary stage. At present, the compression algorithms for interested target areas at home and abroad mainly comprise:
① adopts two region of interest (ROI) coding algorithms proposed in JPEG2000 standard, namely Max-shift and general-scaling-based, the former needs to accurately code the shape of the ROI, which increases the complexity of coding/decoding, and the latter lacks flexibility of defining any offset of the ROI and can not flexibly adjust the contrast of the reconstructed image quality of the ROI and the background area.
② in order to overcome the disadvantages of the first two algorithms, Zhou Wang et al propose an improved algorithm, i.e., Bitplane-by-Bitplane Shift, based on the ideas of the two algorithms, wherein the improved algorithm can realize the coding of an interested region of any shape and can adjust the contrast of the reconstructed image quality of the interested region and the background region, but essentially still popularizes the Max-Shift algorithm.
③ 2003 Lijie Liu et al proposed a new improved translation algorithm (significantplane Shift), which not only has the advantages of the Bitplane-by-Bitplane Shift algorithm, but also can realize the coding of multiple interested regions according to different priorities.
④ Claudio.M proposes six image processing algorithms for extracting local maximum value for obtaining interested region according to human eye visual characteristics, and later, it supplements wavelet transform based algorithm, discrete DCT transform based algorithm, Gaussian filter and Laplace transform based algorithm and 7 x 7 region template matching result statistical algorithm.
⑤ Jerome M.Shapiro proposes an embedded wavelet bit coding algorithm, which mainly comprises an embedded zerotree wavelet transform algorithm (EZW) and a multilevel number set fragmentation algorithm (SPIHT), and the coding stream after the compression processing of the wavelet coefficient of the object of interest is placed at the front end of the whole compressed file and is prior to the coding stream after the compression processing of the wavelet coefficient of the background area.
The invention content is as follows:
the invention provides an unmanned aerial vehicle reconnaissance image compression method which can solve the contradiction between the recovery quality of a target area and the image compression ratio and can meet the requirement of unmanned aerial vehicle reconnaissance to a great extent.
The invention adopts the following technical scheme: an unmanned aerial vehicle scout image compression method comprises the following steps
The value of n is calculated and,
initializing the important coefficient list LSP to be a null list, and enabling all the important coefficient lists to belong to a set RROOT∪RLLThe nodes (i, j) of (a) are stored in the unimportant coefficient list LIP, and all the nodes (i, j) belonging to the set R are stored in the unimportant coefficient list LIPROOTThe node (i, j) of (a) is stored in the unimportant aggregate list LIS and is regarded as a type A element;
step 2, sorting
2.1 calculating the significance function S of all nodes (i, j) in the list LIPn(i,j)
If SnIf (i, j) is 1, adding the node (i, j) into the list LSP, and simultaneously outputting the sign bit of the corresponding wavelet coefficient C (i, j);
2.2 analysis of all nodes (i, j) in the List LIS
A. If node (i, j) is a class A element, its importance function value S is calculatedn(D (i, j)); if Sn(D (i, j)) -1, then:
a. calculating the importance function value S of all nodes (k, l) in the set O (i, j)n(k,l)
If SnIf (k, l) is 1, storing the node (k, l) into the list LSP, and outputting the sign bit of the corresponding wavelet coefficient C (k, l),
if SnIf (k, l) is 0, the node (k, l) is stored at the tail of the list LIP;
b. if the set L (i, j) is an empty set, namely L (i, j) ≠ phi, storing the node (i, j) to the tail of the list LIS, regarding the node (i, j) as a B-type element, and then continuing to perform the processing in the step B; otherwise, deleting the node (i, j) from the list LIS;
B. if node (i, j) is a B-type element, its importance function value S is calculatedn(L (i, j)), if Sn(L (i, j)) > 1, all nodes (k, L) belonging to the set O (i, j) are stored as class a elements to the tail of the LIS, and at the same time, the nodes (i, j) are transferred from the list LIS;
step 3, gradual progress
Calculating absolute values | C (i, j) | of wavelet coefficients C (i, j) corresponding to all nodes (i, j) in the list LSP, and outputting bit values of nth bits of the absolute values;
step 4, updating threshold value
And setting n as n-1, and returning to the step 2 to perform the next-stage coding scanning.
Further, after compression coding is completed, the two parts of compressed data information and the coordinate position information of the region of interest are coded according to a certain data frame structure sequence, and then the image information data are transmitted to a ground receiving station through a wireless channel.
The invention has the following beneficial effects: the unmanned aerial vehicle scout image compression method selects a segmentation compression method, before compression processing, an interested scout target area and a background area of the unmanned aerial vehicle scout image are completely separated and do not influence each other, so that a proper compression method can be freely selected for the two parts. The unmanned aerial vehicle reconnaissance image compression method separately selects different bit rates for the region of interest and the background region, and ensures that the region of interest to be reconnaissance is compressed with a small compression ratio and the background region is compressed with a large compression ratio. Meanwhile, the segmentation compression method can also freely select the size of the compression ratio, and the image quality of the target and the background can be freely adjusted. And the compression bit rate of the reconnaissance target and the background area can be freely adjusted according to the requirement of the transmission bandwidth of the equipment, so that the compression requirement of the reconnaissance image of the unmanned aerial vehicle is met. Meanwhile, the better the recovery quality of the region is, the larger the peak signal-to-noise ratio of the corresponding region is.
Description of the drawings:
fig. 1 is a noisy image.
Fig. 2 is a noise-free image.
Fig. 3 is a closed curve.
Fig. 4 is a diagram of extracting a region of interest.
Fig. 5 is a schematic block diagram of a region-of-interest segmentation-based compression method.
Fig. 6 shows simulation results of the unmanned aerial vehicle scout image compression method of the present invention.
Fig. 7 is a comparison of simulation results with the general displacement method and the SPIHT method.
Fig. 8 is a processing flow chart of the unmanned aerial vehicle scout image compression method.
The specific implementation mode is as follows:
the present invention will be described in further detail with reference to the accompanying drawings and examples.
The unmanned aerial vehicle reconnaissance image is shot by the unmanned aerial vehicle to the ground in the air, and the obtained picture is similar to a picture shot by a person to the ground on an aerial vehicle flying at high altitude. For a person and an airplane, the person can directly obtain related information from the shot picture; for the unmanned aerial vehicle, the shot picture needs to be transmitted to the ground station from the air, and then the relevant information can be acquired manually. Therefore, for the picture shot by the unmanned aerial vehicle, the picture needs to be compressed and then transmitted to the ground, and the information in the picture is acquired after the picture is processed by ground staff.
Because unmanned aerial vehicle flying height is higher, and is fast, and airborne sensor field angle is great, so target pixel is little and the quantity is big in the unmanned aerial vehicle reconnaissance image, and intra-frame correlation is space correlation and other ordinary image difference, and inter-frame correlation is also not strong as time correlation. Some applications of the unmanned aerial vehicle in military are to acquire some military intelligence information from a shot picture, and meanwhile, for the whole image, people are interested in a certain part of the area. In an image of being shot by unmanned aerial vehicle, it can be generally that some military equipment objects have contained military information, say military vehicle, tank, pillbox etc.. Therefore, for unmanned aerial vehicle scout images, people are concerned about military target portions. However, since the unmanned aerial vehicle scout image is shot by the unmanned aerial vehicle at high altitude, military targets such as military vehicles, tanks and pillboxes are very small in picture, that is, in the unmanned aerial vehicle scout image, the military scout target concerned by people only occupies a small partial area of the whole image.
For unmanned aerial vehicle reconnaissance images covering military information, military reconnaissance targets are small and only occupy a small part of area. And the rest large-scale area is natural environment and landscape facilities such as mountains, roads, trees and the like, people are not very concerned about the contents, and the contents are generally used as reference information for subsequently judging the geographical position of the military reconnaissance target. When the unmanned aerial vehicle scout image is actually acquired, people hope that after decompression, the interested scout target part can be clearly identified, and the rest background part is not required to be very high and has a reference value.
For the unmanned aerial vehicle scout image, a compression processing method is hoped to be adopted, so that the interested scout target part of the unmanned aerial vehicle scout image can be reproduced with high quality, and the data volume of the whole image is reduced to a certain extent. The general common compression method is basically to compress the whole image to the same degree, and as the reconnaissance target area in the reconnaissance image of the unmanned aerial vehicle is a small-range area, if the reconnaissance target is required to be reproduced with high quality, the selected compression ratio is limited, and the requirement of greatly reducing the data volume of the image is difficult to meet; if the data amount of the image is greatly reduced, the compression ratio is selected to be large, and it is difficult to ensure high-quality reproduction of the reconnaissance target therein. Therefore, the common compression method is not suitable for the unmanned aerial vehicle scout image and can not meet the requirement of the unmanned aerial vehicle scout image. For the requirement, a compression method for the region of interest is expected, and during processing, the region of interest of the unmanned aerial vehicle scout image is compressed by a small compression ratio or even by a lossless compression ratio, so that high-quality reproduction of the region of interest of the scout image is ensured, and the background region of the region of interest of the unmanned aerial vehicle scout image is compressed by a large compression ratio, so that the data volume of the image is greatly reduced.
In general, a general news image is a large-scale area in which main information is carried, and a drone scout image is a small-scale area in which main information is carried. And adopting a proper compression method according to the compression requirement of the unmanned aerial vehicle scout image. For the unmanned aerial vehicle scout image, the data volume of the whole unmanned aerial vehicle scout image is greatly reduced while the scout target of the unmanned aerial vehicle scout image is required to be clearly reconstructed.
The principle of the region-of-interest segmentation compression method mainly comprises the following aspects:
and I, extracting a region of interest. The invention aims at unmanned aerial vehicle scout images, and the scout targets of the unmanned aerial vehicle scout images are military equipment such as military vehicles, pillboxes, tanks and the like, the targets are regular graphs, and the large-area natural background in the images has no regularity. In the interesting region extraction stage, an interesting target object with specific characteristics in the whole unmanned aerial vehicle scout image needs to be found by using a certain method, and after the object is found, a small region where the object is located is determined to be the interesting region of the unmanned aerial vehicle scout image, so that the interesting scout target region of the unmanned aerial vehicle scout image is extracted.
The extraction of the region of interest mainly comprises Roberts edge detection, Sobel edge detection, Prewitt edge detection, Canny edge detection and the like. Through comparative analysis, the Canny edge detection operator is selected to find the edge contour curves of the image object, and the edge contour curves are processed according to the regularity characteristics of the reconnaissance target graph to determine the reconnaissance target of the reconnaissance image of the unmanned aerial vehicle. The classical Canny edge detection algorithm is a smoothing filter with a gaussian function. The purpose of image smoothing filtering is to improve the signal-to-noise ratio and eliminate noise. However, when the unmanned aerial vehicle scout image is subjected to Gaussian smoothing, edges serving as high-frequency components can be smoothed, and meanwhile, some edges with low intensity can be filtered. Meanwhile, high and low threshold processing is required in the Canny detection process, and edges are connected, so that the selection of the high and low thresholds is also important. And (4) selecting a slightly improved Canny edge detection algorithm to process the unmanned aerial vehicle scout image, and finding out the graph edge contour information of the unmanned aerial vehicle scout image.
① when smoothing filtering the original unmanned aerial vehicle scout image, selecting two-dimensional Gaussian function G (x, y) and its first derivative G' (x, y) to perform convolution operation with the original unmanned aerial vehicle scout image f (x, y). in the actual calculation process, discretizing G (x, y) and recording G (i, j).
P(i,j)＝G(i,j)*f(i,j) (1)
Q(i,j)＝G′(i,j)*f(i,j) (2)
② calculating the amplitude and direction of the gradient, calculating the gradient of the convolution result, and extracting one-dimensional gradient vectors in the x and y directions respectively.
The gradient direction θ (i, j) at the point (i, j) can be obtained from equation (3).
The gradient amplitude M (i, j) at the point (i, j) is obtained from equations (4) and (5).
M(i,j)＝[Qy·sin(θ(i,j))+Qx·cos(θ(i,j))](6)
③ perform non-maximum suppression of gradient amplitudes.
④ double threshold processing and edge connection, the selection of high and low thresholds is important to determine the amount of detected edge information and the continuity of the edge, the local variance reflects the local change condition of the image, and in order to obtain the whole change condition of the image, the average variance of the image is used as one of the high threshold parameters.
Th＝k×(Eave/Fave) (7)
Wherein, ThFor the high threshold sought, fm(i, j) is the gray value of the pixel point (i, j), Eave，Fave，LwAnd LhThe average variance of the image, the average gray value, the width and the height of the image are respectively. Parameter k is (0.2,0.5), low threshold value of Tl＝0.4×Th。
After the high and low thresholds are determined, the image edge contour curves are connected according to the two thresholds.
And (3) carrying out the image processing on the military vehicle by using the improved Canny algorithm, and simultaneously adding salt and pepper noise with the noise density of 0.02 into the original image. The results of the detection are shown in FIGS. 1 and 2 below. From the results of fig. 1 and 2, it can be seen that the improved Canny algorithm avoids the loss of weak edges while maintaining the advantages of the conventional Canny algorithm, and it also has some improvement in noise suppression. When the high and low threshold values are selected, the high and low threshold values are determined according to the mean value and the variance of the image, so that the limitation of manual setting is avoided, and the high and low threshold values can be automatically calculated.
The Canny algorithm before and after improvement is used for processing unmanned aerial vehicle scout images, Lena images and pillbox images, and meanwhile, pepper salt noise with the noise density of 0.02 is added into original images. Firstly, finding an edge contour curve of an unmanned aerial vehicle scout image, after finding the edge contour curve of the image through the slightly improved Canny algorithm, judging the closure of the found edge contour curve by using an eight-neighborhood search method, finally judging whether the found closed contour curve has the regularity characteristic, finally determining that the area where the regular closed curve is located is the interested area of the whole unmanned aerial vehicle scout image, and then determining the closed curve. Closed, as the name implies, i.e., closed, has no gaps. In the invention, the region where the regular military target is located is required to be found, then the target and the background of the image are separated, and then the subsequent compression processing is carried out. After the slightly improved Canny algorithm processing, the edge contour curve of the unmanned aerial vehicle scout image is obtained. In order to find the regular pattern area, it is necessary to find a certain contour curve first, and then determine whether the contour curve encloses a regular pattern. It is therefore necessary to retain only some closed contour curves from all edge contour curves found.
During experimental treatment, the graph edge contour curve image of the unmanned aerial vehicle scout image is subjected to subsequent treatment, a closed curve in the graph edge contour curve image is found according to eight-field information of a determined point, and the specific algorithm is based on that ① a point p (x, y) is set as a starting point A of an image edge0② scanning the 3 × 3 neighborhood of the current pixel point in the counterclockwise direction until a pixel point consistent with the fixed point gray value is encountered, which is the new image edge point An③ subsequent boundary point A when scannednI.e. the start point (closed contour), or the subsequent boundary point anNo connected subsequent boundary point (non-closed contour) is found, the scanning is finished, ④ the obtained closed contour is recorded, the non-closed contour is abandoned, and other edge contours are searched until the image is finished.
However, in the actual processing, since the unmanned aerial vehicle scout image is a picture of the unmanned aerial vehicle taken in the high altitude, the area occupied by the military scout target therein is also relatively small, so that the contour may not be very prominent in the contour detection processing. It may cause that the actually found contour curve is not completely closed, but the military reconnaissance target has a certain size, so that the search algorithm is processed flexibly when searching for the closed curve. When the end point and the start point of a certain curve do not overlap, the curve can be preliminarily determined to be closed when the following expression is satisfied.
In the formula, d is the distance between the end point and the starting point, the distance is calculated according to the coordinates of the pixel points, and sum is the sum of all the pixel points on the curve from the starting point to the end point. Where the range limit 10 for the distance d is a value preselected during the actual treatment. Although the military reconnaissance target of the unmanned aerial vehicle reconnaissance image occupies a smaller area, the edge contour of the military target also has a certain size. In the actual processing, a size range is set, namely the size range of the length, the width or the radius is [20,80 ]. From the size range, it can be seen that the minimum circumference of the military target object profile is 80, and excluding the maximum gap distance of 10, sum in the above formula is at least 70, and thus the ratio of sum to d is at least 7.
The processing results are shown in fig. 3 below.
And then judging the regularity. After the closed curves are obtained, regular graphs in the closed curves can be determined by judging the regularity of the closed curves, so that the region of interest is found. According to the geometric moment-invariant property of the graph, a certain proportional relation exists between the perimeter and the area of the regular graph, so that whether the detected closed contour is the regular graph or not can be judged according to the characteristic.
The moment features mainly characterize the geometric features of the image area. Assuming an image as f (x, y), the (p + q) order geometric moment of f (x, y) is:
Mpq＝∫∫xpyqf(x,y)dxdy,p,q＝0,1,2,…(11)
then the centroid of the object:
center moment:
if the processed image is a discrete image, its geometric moments:
in the binary image, if (i, j) is an object internal point, f (i, j) is 1, and if f (i, j) is 0 for other points. Wherein M is00Representing the area of an object in the binary image.
I.e. the sum of all pixels contained in the closed contour pattern is the area s of the pattern.
Because the geometric invariant moment property of the graph, a certain proportional relation exists between the perimeter and the area of the regular graph, and the proportional formula adopted by the method is l2Where l is the perimeter, i.e. the sum of the perimeter pixels of the closed contour figure, and s is the area of the figure, i.e. the sum of all pixels contained in the closed contour figure. From top to bottom, scanning the read-in closed curve area point by point from left to right, recording the sum l of edge contour pixel points and the sum m of all pixel points in the closed area, wherein l represents the perimeter of the graph, and s (l + m) represents the area of the graph. The proportional calculation formula is:
in actual processing, the read closed curve may not be completely closed, a gap may exist, and the gap needs to be filled before calculating the perimeter and the area of the gap, so that the gap is completely closed. Although the military reconnaissance target only occupies a small area in the whole unmanned aerial vehicle reconnaissance image, the edge contour of the military target also has a certain size. This patent is directed to unmanned aerial vehicle reconnaissance image that contains military reconnaissance target (tank, military vehicle, pillbox), when actually handling this type of unmanned aerial vehicle reconnaissance image, has set for a size scope for the military target wherein, its size scope of length width or radius is [20,80 ]. Under the condition of the size range, for a rectangle, the perimeter range [80,320], the area range [400,6400 ]; for a circle, the perimeter is in the range of [40 π,160 π ], and the area is in the range of [400 π,6400 π ]. The two are combined, the range of the perimeter [80,160 π ], and the range of the area [400,6400 π ]. Therefore, before determining the proportional relationship between the perimeter and the area, the formula (18) is satisfied
In the experimental process, whether the detected closed area is a regular graph or not can be judged according to the r value. In this patent, the user is interested in the military reconnaissance target comparison among the unmanned aerial vehicle reconnaissance image, hopes to find the region that military reconnaissance target belonged to this region is the interesting region of unmanned aerial vehicle reconnaissance image. When 16 ≦ r ≦ 25, it is inferred that the closed contour being processed is a rectangular shape, i.e., corresponding to a military vehicle or tank. When r is 4 pi, the processed closed contour is inferred to be a circular shape, i.e., the corresponding pillbox.
And judging the closed contour graphs in the figure 3, judging whether the graphs are regular graphs or not, keeping the regular closed contours, and abandoning the irregular closed contours. And after the regular area in the image is obtained, according to the coordinate positions of all the pixel points of the selected regular pattern, obtaining the pixel point (minx, miny) formed by the minimum horizontal coordinate and the minimum vertical coordinate and the pixel point (maxx, maxy) formed by the maximum horizontal coordinate and the maximum vertical coordinate. And determining a rectangular region formed by the upper vertex (minx, miny) and the lower vertex (maxx, maxy) as a region of interest to be extracted as a military target region of interest of the unmanned aerial vehicle scout image required by us.
The processing results are shown in fig. 4. As shown in the above figure, fig. 4(a) shows that in the military vehicle image, the target area of the military vehicle of interest is finally determined, and fig. 4(b) shows that in the pillbox image, the target area of the pillbox of interest is finally determined, i.e., the region of interest is determined.
Then, the interesting region and the background region are separated, the interesting military reconnaissance target region in the unmanned aerial vehicle reconnaissance image is found, a rectangular region where the military target is located is determined to be used as the interesting region of the image, namely the interesting region and the background region of the image are determined, and the interesting Region (ROI) and the background region (BG) of the original image need to be separated.
And secondly, separating the images. After finding the reconnaissance target, determining a small-area rectangular range in which the reconnaissance target is located as an interested area R of the reconnaissance image of the unmanned aerial vehicle, and simultaneously recording coordinate position information of the lower interested area. Judging whether each pixel of the unmanned aerial vehicle scout image belongs to a region R or not, and separating the original unmanned aerial vehicle scout image into a region of interest (ROI) and a background region (BG).
The separation method selected by the invention is a separation method based on the region idea.
The regions are separated based on the region idea, namely, some pixel points with similar properties are connected, so that the separation of different regions of the image is realized. The separation during the experimental treatment is described in detail below:
(1) the region of interest is isolated. The method comprises the steps of sequentially scanning an original unmanned aerial vehicle scout image from left to right and from top to bottom, and judging whether each pixel point (x, y) of the image belongs to a region R (wherein R is a rectangular region of interest obtained according to the position information of a detected scout target, the upper vertex of the rectangle is (minx, miny), and the lower vertex of the rectangle is (maxx, maxy).
And judging whether the analysis belongs to R or not for each pixel point (x, y) obtained by scanning. If (x, y) is equal to R, keeping the original gray value f (x, y); otherwise, the reverse is carried outLet its gray value f (x, y) be 0. Scanning is carried out till the end of the picture, and finally the processed image is displayed. I.e. a sub-picture containing only the image region of interest.
(2) The background area is isolated. The processing can be performed in the same manner according to the above steps, only the original gray value of the pixel point in the background region is kept, and the gray value of the pixel point in the region of interest is set to 0. However, since the range of the region of interest is small, that is, the number of pixels to be set to 0 is limited, the above method can be slightly improved, and some calculation amount is reduced.
In this process, the original drone scout image may be scanned only in the region of interest, i.e. from left to right starting from point (minx, miny) and so on from top to bottom. When scanning to point (maxx, miny), we consider wrapping, i.e. starting from point (minx, miny +1) again, and so on until scanning to point (maxx, maxy). In the scanning process, the acquired pixel points are all in the image interesting area, so that the gray value f (x, y) of the acquired pixel points is only required to be set to be 0. And finally, displaying the processed image, namely the subgraph only containing the background area.
Through the separation steps, the separation of the region of interest (ROI) and the background region (BG) of the unmanned aerial vehicle scout image is realized. In the background region subgraph, the region-of-interest information is abandoned, and the background region information is reserved. In the interesting region subgraph, background region information is abandoned, and interesting region information is reserved. After the image is processed by the wavelet transform of N levels, the image has 1 low frequency band and 3N high frequency bands. Wherein wavelet coefficients of high frequency bands are small, and they represent information (detail coefficients) of detailed parts of the image; wavelet coefficients of low frequency bands are relatively large, and they represent information (contour coefficients) of the contour portion of the image. Although wavelet transform does not achieve compression processing on an image by itself, the transformed wavelet coefficients have some characteristics which enable the image to be compressed conveniently.
In the patent, an EZW algorithm and an SPIHT algorithm are two embedded compression coding algorithms commonly used at present based on wavelet transformation, and the EZW algorithm and the SPIHT algorithm both use a successive approximation quantization method, that is, a threshold value of each scanning is half of a last threshold value, and an initialization process and a refinement process of the EZW algorithm and the SPIHT algorithm are similar.
But the construction of the zero tree by SPIHT is more efficient than EZW. For example, the following steps are carried out: assuming that c (i, j) is important and its descendants (including children) are not, in EZW the first scan assigns c (i, j) to the symbol P, forming a four-class tree for the following descendants, denoted by pzz five symbols; in SPIHT, the first scan moves c (i, j) from LIP to LSP and outputs a 1, and after comparing (i, j) in LIS outputs a 0, i.e., represented by two symbols. Therefore, the SPIHT algorithm can better utilize the characteristics of the wavelet coefficients and has better coding performance.
And thirdly, compressing the image. After the unmanned aerial vehicle scout image is subjected to segmentation processing, the unmanned aerial vehicle scout image is divided into two sub-images, and the two sub-images can be compressed by different compression ratios respectively. The interested area is compressed by low compression ratio or even lossless compression, the background area is compressed by high compression ratio, and different compression methods can be selected to compress the background area according to the requirement of transmission bandwidth. This ensures both a high efficiency compression of the image and a high quality reproduction of the region of interest.
Through the analysis, the SPIHT algorithm is selected to compress the reconnaissance target area and the background area of the reconnaissance image of the unmanned aerial vehicle. And selecting SPIHT with different bit rates to process the two subgraphs according to different compression requirements of the two parts. Different bit rates correspond to different compression ratios, the larger the bit rate, the smaller the compression ratio.
After the above operation processing, the original unmanned aerial vehicle scout image is separated into a region of interest (ROI) only containing scout target information and a background region (BG) only containing background information, and the conventional SPIHT algorithm scans codes from beginning to end, which causes waste of code streams. The SPIHT algorithm is improved properly according to the characteristic that the two parts of subgraphs contain information. The basic idea is as follows: for subgraphs containing only ROI coefficients, those points and sets located in BG are not encoded. For subgraphs containing only BG coefficients, those points and sets located in the ROI are not encoded.
Taking the interesting region subgraph only containing the military reconnaissance target information as an example, the compression processing flow is described, and the analogy can be given to the background region subgraph only containing the background information.
1. Initialization
The value of n is calculated and,
initializing the important coefficient list LSP to be a null list, and enabling all the important coefficient lists to belong to a set RROOT∪RLLThe nodes (i, j) of (a) are stored in the unimportant coefficient list LIP, and all the nodes (i, j) belonging to the set R are stored in the unimportant coefficient list LIPROOTIs stored into the unimportant aggregation list LIS and is considered as a class a element.
2. Sorting
(1) Calculating the significance function S of all nodes (i, j) in the list LIPn(i,j)。
If SnAnd (i, j) ═ 1, adding the node (i, j) into the list LSP, and simultaneously outputting the sign bit of the corresponding wavelet coefficient C (i, j).
(2) All nodes (i, j) in the list LIS are analyzed.
A. If node (i, j) is a class A element, its importance function value S is calculatedn(D (i, j)). If Sn(D (i, j)) -1, then:
a. calculating the importance function value S of all nodes (k, l) in the set O (i, j)n(k,l)。
If SnIf (k, l) is 1, the reaction will proceedAnd storing the nodes (k, l) into the list LSP, and simultaneously outputting the sign bit of the corresponding wavelet coefficient C (k, l).
If SnIf (k, l) ═ 0, the node (k, l) is stored at the tail of the list LIP.
b. If the set L (i, j) is an empty set, namely L (i, j) ≠ phi, storing the node (i, j) to the tail of the list LIS, regarding the node (i, j) as a B-type element, and then continuing to perform the processing in the step B; otherwise, the node (i, j) is deleted from the list LIS.
B. If node (i, j) is a B-type element, its importance function value S is calculatedn(L (i, j)). If SnIf (L (i, j)) > 1, all nodes (k, L) belonging to the set O (i, j) are stored as class a elements at the end of the LIS, and the nodes (i, j) are moved out of the list LIS.
3. Progression of motion
And calculating absolute values | C (i, j) | of wavelet coefficients C (i, j) corresponding to all nodes (i, j) in the list LSP, and outputting the bit value of the nth bit.
4. Threshold update
And setting n as n-1, and returning to the step 2 to perform the next-stage coding scanning. Bit rate (R) of the region of interest during the experimental treatment in order to make the region of interest quality better than the background region qualityroi) Bit rate (R) to be higher than background regionbg). After compression coding is finished, the compressed data information of the two parts and the coordinate position information of the interested area are coded according to a certain data frame structure sequence, and then the image information data are transmitted to a ground receiving station through a wireless channel. After receiving image data information, firstly carrying out SPIHT decoding and wavelet inverse transformation of corresponding coding bit rate on the received information, wherein the decoding process is the inverse process of coding, and correspondingly inlaying the decompressed interesting region subgraph to the relative position of the decompressed background region subgraph according to the received coordinate position information of the interesting region, thereby reconstructing an original image.
And fourthly, transmitting or storing. After compression, the two parts of compressed data information and the coordinate position information of the region of interest are combined together and transmitted to a ground receiving station through a wireless channel.
And fifthly, reconstructing the image. After the image data information is received, decompression operation is carried out, an interesting region sub-image and a background region sub-image are synthesized according to the coordinate position of the interesting region, and an original unmanned aerial vehicle reconnaissance image is reconstructed.
In general, a simulation tool is selected to firstly simulate and test the processing effect of the algorithm, the difference between the obtained result and the expected effect is analyzed, then the algorithm program is debugged, and the performance of the algorithm is gradually improved, so that the better processing effect is realized, and the expected value is reached.
The method is used for carrying out simulation debugging and verification on a newly-proposed segmentation compression algorithm based on the region of interest under the MATLAB programming environment. In the experimental process, an unmanned aerial vehicle reconnaissance image is selected as an experimental image. However, the original unmanned aerial vehicle reconnaissance image is large and is not easy to display on a screen, so that only one region is selected for experiment simulation processing in order to facilitate use in the experiment process, and the resolution of the processed image in the experiment is not small.
The maxshift method in the JPEG2000 standard is to scale up or shift up wavelet coefficients associated with an object of interest after the wavelet transform process, so that the wavelet coefficients are located at a bit plane higher than the wavelet coefficients of the background portion of the image. The general SPIHT is an embedded wavelet bit encoding mode, and an encoding stream after the compression processing of the wavelet coefficient of an interested target is placed at the front end of the whole compressed file and is prior to the encoding stream after the compression processing of the wavelet coefficient of a background area. The wavelet coefficients related to the ROI are all moved upwards to enable the bit plane where the wavelet coefficients are located to be higher than the bit plane where the background regional coefficients are located, then in the subsequent embedded coding, the wavelet coefficients of the ROI are located in front of the wavelet transform coefficients of the BG, and therefore the ROI is coded and refined before the BG.
The segmentation compression algorithm based on the region of interest selected by the patent aims to solve the contradiction between the recovery quality of the target region and the image compression ratio and achieve real-time transmission of the image.
In the experimental simulation process, two parts of the image are decompressed by different bit rates, namely, the two parts correspond to different compression ratios. In order to ensure that the image restoration quality in the small region range of the reconnaissance target is higher than that of the background region, the bit rate of the region of interest is selected to be higher than that of the background region.
This patent also can follow the angle of ration to analyze and use behind the above-mentioned method, the recovery quality in the target area of the reconnaissance of unmanned aerial vehicle spy image's interest reconnaissance and background area, and it is no longer repeated here.
In fig. 7, a graph (a) is a grayscale graph of an original military vehicle image, a graph (b) is a result of a simulation process by a general displacement method, a graph (c) is a result of a simulation process by an SPIHT algorithm, and both a graph (d) and a graph (e) are results of a process of selecting the segmentation compression method of the present patent. Graph (b) data volume: 15108, data amount in graph (c): 15062.
the graph (d) ensures that the reconstruction quality of the reconnaissance target area of the reconnaissance image of the unmanned aerial vehicle is equivalent to that of the graphs (b) and (c), and the compression amount of the background area is larger than that of the graphs (b) and (c). Therefore, when the reconnaissance target of the reconnaissance image of the unmanned aerial vehicle is reconstructed to the same degree, the data size processed by the separation compression method is smaller than the data size processed by the general displacement method and the SPIHT algorithm, namely, the data size is reduced. Graph (d) data volume: 5148.
the image (e) is to make the reconstruction quality of the target region of the unmanned aerial vehicle scout image higher than that of the images (b) and (c), i.e. the ROI region bit rate of the image (e) is selected to be higher than that of the image (d), and at the same time, the compression amount of the background region is increased, i.e. the BG region bit rate of the image (e) is lower than that of the image (d). Therefore, the military reconnaissance target can be seen clearly, and the data volume of the image is reduced. Graph (e) data volume: 4826.
compared in a comprehensive mode, the unmanned aerial vehicle reconnaissance image compression method is a separation compression method, before compression processing, the interesting reconnaissance target area and the background area of the unmanned aerial vehicle reconnaissance image are completely separated, and the interesting reconnaissance target area and the background area are not affected with each other, so that a proper compression method can be freely selected for the two parts. According to the method, different bit rates are separately selected for the interesting region and the background region, so that the small compression ratio compression of the interesting reconnaissance target region is ensured, and the large compression ratio compression of the background region is ensured. Meanwhile, the separation compression method can also freely select the size of the compression ratio, and the image quality of the image target and the background can be freely adjusted. And the compression bit rate of the reconnaissance target and the background area can be freely adjusted according to the requirement of the transmission bandwidth of the equipment, so that the compression requirement of the reconnaissance image of the unmanned aerial vehicle is met. Meanwhile, the better the recovery quality of the region is, the larger the peak signal-to-noise ratio of the corresponding region is.
The foregoing is only a preferred embodiment of this invention and it should be noted that modifications can be made by those skilled in the art without departing from the principle of the invention and these modifications should also be considered as the protection scope of the invention.
Claims (1)
1. An unmanned aerial vehicle scout image compression method is characterized in that: comprises the following steps
Detection of an object of interest
Finding edge contour curves of the image object by using a Canny edge detection operator, processing the edge contour curves according to the regularity characteristics of the reconnaissance target graph, and determining the reconnaissance target of the reconnaissance image of the unmanned aerial vehicle;
① smoothing the original unmanned aerial vehicle scout image, selecting two-dimensional Gaussian function G (x, y) and its first derivative G' (x, y) to perform convolution operation with the original unmanned aerial vehicle scout image f (x, y), discretizing G (x, y), and recording as G (i, j)
P(i,j)＝G(i,j)*f(i,j) (1)
Q(i,j)＝G′(i,j)*f(i,j) (2)
② calculating gradient amplitude and direction, calculating gradient of the convolution result, and extracting one-dimensional gradient vector in x and y directions
From the equation (3), the gradient direction θ (i, j) at the point (i, j) can be obtained
Obtaining a gradient amplitude M (i, j) at the point (i, j) from the equations (4) and (5)
M(i,j)＝[Qy·sin(θ(i,j))+Qx·cos(θ(i,j))](6)
③ performing non-maximum suppression on gradient amplitudes;
④ Dual thresholding and edge joining
Th＝k×(Eave/Fave) (7)
Wherein, ThFor the high threshold sought, fm(i, j) is the gray value of the pixel point (i, j), Eave，Fave，LwAnd LhRespectively mean variance, mean gray value, width and height of the image, parameter k is (0.2,0.5), and low threshold value is selected Tl＝0.4×Th；
Second, separation of target and background
After finding the reconnaissance target, determining a small-area rectangular range in which the reconnaissance target is located as an interested area R of the reconnaissance image of the unmanned aerial vehicle, recording coordinate position information of the lower interested area, judging whether each pixel of the reconnaissance image of the unmanned aerial vehicle belongs to the area R, and separating the original reconnaissance image of the unmanned aerial vehicle into an interested area (ROI) and a background area (BG);
(1) separating out an interested region, sequentially scanning an original unmanned aerial vehicle scout image from left to right and from top to bottom, and judging whether each pixel point (x, y) of the image belongs to a region R, wherein R is a rectangular interested region obtained according to the detected position information of a scout target, the upper vertex of the rectangle is (minx, miny), and the lower vertex of the rectangle is (maxx, maxy);
judging whether the analysis belongs to R or not for each pixel point (x, y) obtained by scanning, and if (x, y) belongs to R, keeping the original gray value f (x, y); otherwise, the reverse is carried out
The gray value f (x, y) is set to 0, the scanning is carried out until the picture is finished, and finally the processed image is displayed, namely the sub-image only containing the image interesting region
(2) Separating out a background area, scanning from left to right from points (minx, miny), scanning from top to bottom, when scanning the points (maxx, miny), changing lines, namely starting from the points (minx, miny +1), and so on until scanning the points (maxx, maxy), setting the gray values f (x, y) of the acquired pixels to 0 only as the acquired pixels are in the image interesting area in the scanning process, and finally displaying the processed image, namely a sub-image only comprising the background area
Through the separation steps, the separation of a region of interest (ROI) and a background region (BG) of the unmanned aerial vehicle scout image is realized;
third, compression coding of image
Step 1, initialization
The value of n is calculated and,
importance of initializationThe number list LSP is a null list, all belonging to the set RROOT∪RLLThe nodes (i, j) of (a) are stored in the unimportant coefficient list LIP, and all the nodes (i, j) belonging to the set R are stored in the unimportant coefficient list LIPROOTThe nodes (i, j) of (a) are stored into an unimportant set list LIS and are regarded as class a elements, wherein LSP is an important coefficient table, LIP is an unimportant coefficient table, and LIS is an unimportant set table;
step 2, sorting
2.1 calculating the significance function S of all nodes (i, j) in the list LIPn(i,j)
If SnIf (i, j) is 1, adding the node (i, j) into the list LSP, and simultaneously outputting the sign bit of the corresponding wavelet coefficient C (i, j);
2.2 analysis of all nodes (i, j) in the List LIS
A. If node (i, j) is a class A element, its importance function value S is calculatedn(D (i, j)); if Sn(D (i, j)) -1, then:
a. calculating the importance function value S of all nodes (k, l) in the set O (i, j)n(k,l)
If SnIf (k, l) is 1, storing the node (k, l) into the list LSP, and outputting the sign bit of the corresponding wavelet coefficient C (k, l),
if SnIf (k, l) is 0, the node (k, l) is stored at the tail of the list LIP;
b. if the set L (i, j) is an empty set, namely L (i, j) ≠ phi, storing the node (i, j) to the tail of the list LIS, regarding the node (i, j) as a B-type element, and then continuing to perform the processing in the step B; otherwise, deleting the node (i, j) from the list LIS;
B. if node (i, j) is a B-type element, its importance function value S is calculatedn(L (i, j)), if Sn(L (i, j)) > 1, all nodes (k, L) belonging to the set O (i, j) are stored as class a elements to the tail of the LIS, and at the same time, the nodes (i, j) are transferred from the list LIS;
step 3, gradual progress
Calculating absolute values | C (i, j) | of wavelet coefficients C (i, j) corresponding to all nodes (i, j) in the list LSP, and outputting bit values of nth bits of the absolute values;
step 4, updating threshold value
Setting n as n-1, and returning to the step 2 to perform the next-stage coding scanning; wherein: c (i, j): wavelet transform coefficients; o (i, j) the set of coordinates for all children of node (i, j); d (i, j) the coordinate set of all the descendants of the node (i, j); importance function Sn(i, j): finger Sn(X), wherein (i, j) is ∈ X;
fourth, data transmission
After compression, combining the compressed data information of the two parts and the coordinate position information of the region of interest together, packaging, and transmitting to a ground receiving station through a wireless channel;
fifthly, reconstructing the image
After the image data information is received, decompression operation is carried out, an interesting region sub-image and a background region sub-image are synthesized according to the coordinate position of the interesting region, and an original unmanned aerial vehicle reconnaissance image is reconstructed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510102937.4A CN104754340B (en) | 2015-03-09 | 2015-03-09 | Unmanned aerial vehicle reconnaissance image compression method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510102937.4A CN104754340B (en) | 2015-03-09 | 2015-03-09 | Unmanned aerial vehicle reconnaissance image compression method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104754340A CN104754340A (en) | 2015-07-01 |
| CN104754340B true CN104754340B (en) | 2020-02-21 |
Family
ID=53593361
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510102937.4A Active CN104754340B (en) | 2015-03-09 | 2015-03-09 | Unmanned aerial vehicle reconnaissance image compression method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104754340B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105721872A (en) * | 2016-01-29 | 2016-06-29 | 广东欧珀移动通信有限公司 | Image compression method, and terminal |
| CN105812814B (en) * | 2016-03-24 | 2018-12-28 | 湖南优象科技有限公司 | A kind of flight data visualization coding and transmission method based on UAV Video |
| CN108933944A (en) * | 2018-06-08 | 2018-12-04 | 玖芯半导体有限公司 | The method for compressing image that a kind of pair of image data is handled |
| CN109005421A (en) * | 2018-08-17 | 2018-12-14 | 青岛海信电器股份有限公司 | Image processing method and device, computer readable storage medium |
| CN109829928A (en) * | 2018-12-29 | 2019-05-31 | 芜湖哈特机器人产业技术研究院有限公司 | A kind of extracting method of target image extracting method and picture position feature |
| WO2020243906A1 (en) * | 2019-06-04 | 2020-12-10 | SZ DJI Technology Co., Ltd. | Method, device, and storage medium for encoding video data base on regions of interests |
| CN111031325A (en) * | 2019-10-28 | 2020-04-17 | 西安万像电子科技有限公司 | Data processing method and system |
| WO2022061880A1 (en) * | 2020-09-28 | 2022-03-31 | 深圳市大疆创新科技有限公司 | Movable platform and online data processing method and apparatus therefor, and storage medium |
| CN112839228B (en) * | 2020-12-21 | 2023-10-20 | 福建智达力胜电力科技有限公司 | Unmanned aerial vehicle image processing and transmitting method |
| CN112967290A (en) * | 2021-02-22 | 2021-06-15 | 中国人民解放军空军航空大学 | Method for automatically identifying enemies of target aircraft in air by unmanned aerial vehicle |
| CN113691305B (en) * | 2021-10-26 | 2022-02-11 | 武汉兴图新科电子股份有限公司 | Cloud-edge-end-based picture transmission method for Beidou third-number short message channel |
| CN114531599B (en) * | 2022-04-25 | 2022-06-21 | 中国医学科学院阜外医院深圳医院(深圳市孙逸仙心血管医院) | Image compression method for medical image storage |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6407990B1 (en) * | 1999-11-11 | 2002-06-18 | Huawei Technologies Co., Ltd. | Method of transmission of image by CDMA system |
| CN102270241A (en) * | 2011-08-16 | 2011-12-07 | 浙江大学 | Image retrieving method based on sparse nonnegative matrix factorization |
| CN102637302A (en) * | 2011-10-24 | 2012-08-15 | 北京航空航天大学 | Image coding method |
| CN103281534A (en) * | 2013-05-15 | 2013-09-04 | 西北工业大学 | Hyper-spectral image compression and encryption method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2003018557A (en) * | 2001-07-04 | 2003-01-17 | Monolith Co Ltd | Method and apparatus for processing image |
-
2015
- 2015-03-09 CN CN201510102937.4A patent/CN104754340B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6407990B1 (en) * | 1999-11-11 | 2002-06-18 | Huawei Technologies Co., Ltd. | Method of transmission of image by CDMA system |
| CN102270241A (en) * | 2011-08-16 | 2011-12-07 | 浙江大学 | Image retrieving method based on sparse nonnegative matrix factorization |
| CN102637302A (en) * | 2011-10-24 | 2012-08-15 | 北京航空航天大学 | Image coding method |
| CN103281534A (en) * | 2013-05-15 | 2013-09-04 | 西北工业大学 | Hyper-spectral image compression and encryption method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104754340A (en) | 2015-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104754340B (en) | Unmanned aerial vehicle reconnaissance image compression method | |
| CN1711551B (en) | Region-of-interest tracking method and device for wavelet-based video coding | |
| JP2003230012A (en) | Image processing based on header information | |
| CN103826125B (en) | Concentration analysis method and device for compression monitor video | |
| US10708601B2 (en) | Methods and apparatuses for encoding and decoding superpixel borders | |
| US20210304357A1 (en) | Method and system for video processing based on spatial or temporal importance | |
| CN103606136B (en) | Based on the video super resolution of key frame and non local constraint | |
| TWI250423B (en) | Method for processing video images | |
| Agustsson et al. | Extreme learned image compression with gans | |
| US20150326878A1 (en) | Selective perceptual masking via scale separation in the spatial and temporal domains using intrinsic images for use in data compression | |
| CN105825491A (en) | Image fusion method based on hybrid model | |
| CN102769749B (en) | Post-processing method for depth image | |
| CN107301662A (en) | Compression restoration methods, device, equipment and the storage medium of depth image | |
| EP3167428A1 (en) | Method and apparatus for up-scaling an image | |
| US20150169632A1 (en) | Method and apparatus for image processing and computer readable medium | |
| CN103985102A (en) | Image processing method and system | |
| Yuan et al. | Object shape approximation and contour adaptive depth image coding for virtual view synthesis | |
| CN103544717B (en) | A kind of two-stage three dimensional image processing coded method based on SIFT feature | |
| US8897378B2 (en) | Selective perceptual masking via scale separation in the spatial and temporal domains using intrinsic images for use in data compression | |
| CN104394399B (en) | Three limit filtering methods of deep video coding | |
| Liu et al. | Improved sparse 3D transform-domain collaborative filter for screen content image denoising | |
| TWI226189B (en) | Method for automatically detecting region of interest in the image | |
| CN103997653A (en) | Depth video encoding method based on edges and oriented toward virtual visual rendering | |
| Kang et al. | Self-learning-based low-quality single image super-resolution | |
| CN104883558A (en) | K-means clustering based depth image encoding method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |