TW390097B - Image compression system with selectable compression ratio - Google Patents

Abstract

The present invention relates to an image compression system with selectable compression ratio, which uses a method for detecting the complexity of images to adjust the quantization pitch of a quantizer in JPEG thereby achieving the objective of balancing between the image quality and adjusting the compression ratio. The system, in its final step, uses a structure of a high frequency coefficient, that is not sharp to the human eyes in discarded discrete cosine transform (DCT) coefficients, to more accurately limit the compression ratio of the image. This image compression system with selectable compression ratio can be used in a device that has a constant storage space, such as a digital camera.

Description

Printed by the Central Standards Bureau of the Ministry of Economic Affairs and Consumer Cooperatives " A7 ------- B7_ V. Description of the Invention (I) This case is an image compression system with selectable compression ratio, especially for digital cameras with fixed Equipment system with large storage space. The purpose of image compression is to reduce the amount of redundant data in the image so that more images can be stored in a limited memory. Recently, in the popular JPEG (Joint Phot (^ graphic Experts Grup, JpEG) image compression system, if different images are compressed with the same quantization table, different images will be generated. Compression ratio. In other words, for more complex images, the compression is lower, and compressed image data will require larger storage space. Conversely, for more monotonous images, the compression is higher, and the compressed image data The amount is relatively small. In digital camera systems, because the size of the memory ceremony is fixed, for example, for a digital camera that can take 32 full-color (true c〇1〇r) photos, the memory size is 2M bytes. For example, the average storage space available for each photo is only 62.5K bytes. Therefore, for a system such as a digital camera, each image is stored with a fixed size of memory for storage. Different complexity causes the image data output after compression is too large to match the storage space. Therefore, images of different complexity must be It is necessary to control its compression ratio to ensure that the compressed image data is of a suitable size. In addition, it must be weighed. After the image data is restored, the image loss should not be too serious. In conventional technology, in order to solve the above problems, usually It takes a lot of calculations to control the compression ratio, and the image format it produces does not fully comply with the JPEG standard, resulting in poor compatibility. __ 2 (d ^ j A4W ^ (210X297 ^) IIO clothing ------ IT ------ ό——I --- (Please read the precautions on the back before filling in this page) Printed by the Consumers ’Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs A7 ____B7____ 5. Description of the invention (〉) For example, the method mentioned in the Filmless Digital Camera With Selective Image Compression method is to divide the image into multiple blocks. With DCT as the basic structure, compression is performed first. When the quality of the reconstructed image is too poor, the block is then Compressed by differential coding method, and finally stored in a digital camera system with pre-configured memory space. The method proposed by Quantization Step Size Adjusting Circuit Using Edge Detections, It is a system similar to adjusting the size of the quantization interval by checking the amount of Edge components in the image, and controlling the compression ratio. However, the adjustment of the quantization interval is based on blocks, and the detection method is different from this case. In addition, the method mentioned in the Image Compression System by Setting Fixed Bit Rates is also a method of detecting the degree of change in the content of the image and adjusting the quantization interval in units of blocks, and detecting the change in the content of the image. The techniques used in the degree are also different from the present case. Although each of the three methods mentioned above has good picture quality, it is not easy to implement, which results in higher costs and higher prices, which are its major drawbacks. Secondly, all three are non-standard JPEGs, which often cause poor compatibility problems in use. In 1974, KRRao published "Discrete Cosine Transform" and other related literatures in IEEE Trans, on Computers, Vol. 23, pp. 90_93, and confirmed that the correlation coefficients P of general images are mostly around 0.95, and when At around 0.95, the DCT system is approaching the best conversion KLT (Karhunen-Loeve Transform, KLT). Due to the characteristics of DCT's fast calculation, the 3 papers are "^ < eNS 丨 ί 2 丨 0X297 mm) ~~ 'I-TIJ -----—— (Please read the notes on the back before filling in this page), 1Τ -Γ A7 B7 Printed by the Consumers' Cooperative of the Central Procurement Bureau of the Ministry of Economic Affairs 5. Instructions for Invention (Order) Regardless of recent JPEG, H.261 and MPEG have adopted DCT as part of their compression system. In 1981, AK Jain mentioned in IEEE Trans. Commun · Vol. COM-29, no. 12. pp 1799-1808, Dec. "Displacement Measurement and Its Allpication in Interframe Image Coding",-Autocorrelation of general image The function can be approximated as: = ..... (1) where ^ and outer represent the correlation coefficients between the horizontal and vertical shirt image data in a single picture, and m and η are the horizontal and vertical distances, respectively. If A = outer = ρ, therefore, using equations (1) and (2), we can predict the approximate correlation between some neighboring pixel data in the spatial domain. In this case, the degree of correlation between pixel correlations is used. To determine the complexity of the image content, the smaller the correlation, the more complex the image, and the larger the quantization interval of the image, the greater the correlation, the more monotonous the image, and the smaller the image spacing; The compression ratio of the image can be controlled. For this reason, in view of the lack of known technology, this creation has been carefully tested and researched, and has persevered in the spirit, and finally created the "Compression Selection of Compression in the Case" System ". Following This case is a brief description of the case. The purpose of this case is to solve the problem of controlling the size of the f image shrinkage data and the storage space in the digital camera system. Furthermore, this case is to maintain the acceptable quality of the reconstructed image. Propose an actual (please read the notes on the back before filling this page)

• IT

A7 B7 Printed by the Central Standards Bureau of the Ministry of Economic Affairs and printed by a separate consumer cooperative. V. Invention Description (y) A technology that can adjust the image compression ratio so that the amount of compressed data is within the acceptable storage space. This case is an image compression system with a selectable compression ratio, which includes a complexity discriminator, whose input is an image data and the output is a three-dimensional table, and a sampling analysis is performed according to the image data to determine its repetition and the quantification. Form;-an encoder that inputs-image data to the complexity scaler, and outputs an image compression data based on the quantized form of the ribbed image; and-a data quantity checker, which is electrically connected to the encoder, and The size of the compressed data of the image output by clamping the encoder does not exceed a predetermined range. 'The image compression system with a selectable compression ratio as described above, wherein the heterogeneity discriminator is a package:-a sampling subtracter, which takes a pixel at a fixed interval of the image data to calculate the absolute sampling of the pixel Sampling Absoule Difference (SAD) positive and negative numbers are electrically connected to the sampling subtracter to determine the value of the pixel = the difference between positive and negative 値; an accumulator is electrically connected to the positive and negative discriminator To accumulate the positive absolute difference of the sampling absolute difference of the pixel; and a subtractor, which is electrically connected to the positive and negative discriminator to add the absolute absolute difference of the sampling absolute difference of the pixel to the absolute value of the negative; and a quantization table determines A quantizer that is electrically connected to the subtractor and uses the result of the calculated absolute difference 取样 as a basis for selecting a quantization table, and outputs a selected quantization table for use by the encoder. The image compression system with selectable compression ratio, wherein the data amount detector includes: a data size calculator to calculate the size of the image compression data, and to calculate the image compression The size of the material does not exceed a limit. This paper size is applicable to the Chinese National Standard (CNS) A4 specification (2 丨 0X297 mm) (please read the precautions on the back before filling in this page ", -β tl Staff of the Central Standards Bureau The consumer cooperative prints A7 _ _B7__ V. When the invention is explained, the image compression data is output; and a high-frequency coefficient deleter is electrically connected to the data size calculator for the size of the image compression data When the limit is exceeded, some high-frequency portions of the image compression data are deleted, so that the size of the image data does not exceed the limit. As described in the image compression system with a selectable compression rate, the encoder is a standard JPEG encoder. The above-mentioned standard JPEG encoder, high-frequency coefficient deleter, data size calculator, subtractor, positive-negative number discriminator, accumulator and other conventional techniques can be implemented by those familiar with the technology, so This case does not go into details. Those who want to elaborate are the architecture and principle of an image compression system with a selectable compression rate. This case can be illustrated by the following diagrams and detailed descriptions. Get a deeper understanding: First picture: The preferred embodiment of the image compression system with selectable compression rate applied to the digital camera system. Second picture: The block diagram of the image compression system encoder with selectable compression rate in this case. The third graph to the eighth graph: Comparison of MAD and SMAD of various complexity. The ninth graph to the twelfth graph: one of the verification results. The thirteenth graph to the sixteenth graph: the second verification result. The above graph The main components of the formula are as follows: 100: lens. 200: electrical coupling element sensor. 300: correlated double sampling / automatic gain controller. 400: analog to digital converter. 6 This paper standard applies to China National Standard (CNS) A4 specification (210X297 mm) (Please read the precautions on the back before filling this page) Order

Printed by the Consumers' Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs. 5. Description of the Invention (厶) 500: Digital signal processor for electrical switching components. 600: JPEG encoder. 610: Standard JPEG encoder. 620: complexity discriminator. 630: Data quantity checker. 631: Data size calculator. 70 0: Memory. 800: Electrically-coupled component clock-driven "chip. This case is mainly a system that detects the complexity of the input image in conjunction with the JPEG system. This system can control the size of the output data after JPEG compression encoding to facilitate the storage space of digital cameras. The fixed system stores the image data. Please refer to the first figure. The first figure is a basic block diagram of a digital camera system. Among them, the lens 100 and the Charge Coupled Device (CCD) 200 are used to convert the optical image signal. Converted into a charge signal. Correlate Double sampling (CDS) circuit and Automatic Gain Control (AGC) circuit 300 are used to improve the low frequency noise output from the CCD and adjust the signal level. The analog digital converter 400, even if the output of the automatic gain control circuit 300, is converted into a digital signal that is convenient for subsequent processing, so as to input the next Charged Coupled Device Digital Signal Processor (CCDSP) 500 0 In CCDSP500, the input signal will continue to be white balanced and color corrected (col or correction), 7 (please read the precautions on the back before filling in this page) l, -β Dimensions of this paper are applicable to China National Standards (CNS) A4 specifications (210X297 mm) A7 ____B7_ V. Description of the invention (^) Gamma correction and other processing to obtain a color signal that can be accurately used. Finally, the JPEG encoder 600 placed after the CCDSP 500 is used to compress the image signal and store it in the memory 700. The device will be described in detail later. The charge_synchronous device clock drive chip 800 is used to provide the system clock. Further explanation, the jPEG encoder 600 in the second figure includes: standard JPEG encoder 610, complexity discriminator 620, and data volume checker 630. ^ Among them, DCT611 in the standard jpeg encoder 610 performs DCT calculation on the input image data to compress the data volume. To the low frequency so that the next quantizer 612 performs the quantization operation, so that the low frequency coefficients that retain more information in the conversion domain are assigned more bit coding. The entropy encoder 613 converts the aforementioned D CT and quantized data are reprocessed, that is, a new representative symbol is assigned to each data, that is, a shorter code is given a higher probability, and a longer code is given if there is a lower probability. Actually reduce the overall data volume and reduce the size of the output data after the entire encoding. The complexity discriminator 62 in the second figure is used to detect the complexity of the image. It can provide different quantizers of the quantizer to control the compression ratio of the image according to the images of different complexity. Printed by the Consumer Affairs Cooperative of the Central Bureau of Standards, Ministry of Economic Affairs (please read the notes on the back before filling out this page). In the previous discussion, we can know that the complexity of the image can be judged by the difference between the correlation between the pixels. However, in actual operation, calculating the correlation coefficient is a rather complicated task. In order to reduce the computation, the method of calculating the Mean Absolute Diiference (MAD) can be used to determine the complexity of the image. The expression is as follows: 8

V. Description of the invention (?) A7 B7 Printed by the Shellfish Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs, MN Nishi-Lion-you can be less you) _F Qiao Shao Nen • Jia_Mom Shao + Lat ....... (3) where 'M and N represent the width and height of the image, respectively, and ργ (χ, γ) represents the number of the brightness signal Y of the π image at the positions x and y. The larger the MAD, the greater the difference between pixels and the more complex the image content. From the second graph to the eighth graph, the results of the experiment are shown. It can be known that the complexity of the image can be divided into six types from the third graph to the eighth by the calculation of the actual MAD. These images are all 640x480. To further explain, although the correlation between general image pixels is different, the correlation is still very high, as the aforementioned wide = 0.95, so if you do j} AD for each pixel As a result, although a more accurate judgment can be obtained, the amount of calculation is relatively increased. In order to reduce the amount of calculation and speed up the execution, you can use the following formula (4) with spatial sampling, and assume that MAD operations are performed only in the horizontal direction, which is called Samplin.g Mean Absoule Difference (SMAD) The following formula: SMAD = + {Sx'i), b + (^ ';,)) ~ FAa + ^' 0 'lb + ^ j)) + \ FY (a + (Sx · i), b + (Sy ·;)) -FY (a + (Sx · 〇 + lb + (Sy · y)) | .. (4) where a and b represent the horizontal and vertical displacements of the starting sampling point, respectively. Between 2 and 10 ° Sx and Sy represent the sampling interval in the horizontal and vertical directions, respectively, and their numbers are generally between 2 and 16. And m = integer 〔(Mac) / Sx], n = integer [ (Nb-

MxNx4i n-

I

I order r% This paper size is applicable to the Chinese National Standard (CNS) A4 specification (210X297 mm) Printed on the A7 B7 by the Shellfish Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs 5. Description of the invention (^) (〇 / Sy) 'Μ, N is the width and height of the image, c and d respectively indicate the end sampling point, the horizontal and vertical displacement 値, the number 値 is generally between 2 and 10. The operation symbol integer [] means that the decimal is rounded off unconditionally, and only integers are taken. The third to eighth pictures also show that when Sx = Sy = 9, a = 6, b = 3 'c = 4' d = 9, that is, the horizontal position is from the 6th pixel, and the vertical position is from the 3rd From the beginning of the pixel, take out 1 every 9 pixels horizontally and vertically. • Dots' to the 636th pixel horizontally and to the 471th pixel vertical. After doing SMAD and MAD operations, compare the two results. From the third figure to the first The experimental results of Figure 8 show that the accuracy of 31 ^ 1) and ether 1) for determining the complexity of general images is very close, but SMAD has the advantage of greatly reducing the amount of calculations. 0 Furthermore, in order to reduce the actual operation more Process, the above calculation of SMAD can be modified to sample absolute difference SAD (Sampling Absoule Difference (SAD), whose operation is as follows: (5) SAD = Σ Σ (kr + (Sx · i), b + (Sy .;)) -FY (a + (Sx + (Sy · j)) \

/ -I; = I + \ FY {a + (Sx · i), b + (Sy · j))-Fr {a + (Sx-/) + 1,6 + (Sy y)) |) = ΣΣ + + (Sy ·;)) -FY (a + (Sx i)-\, b + (Sy · j)) / «I y = i + Fr (a + (Sx i), b + (Sy · j )) -Fr (a + (Sx-/) + 1,6 + (Sy · j))) ^ ~ (FY (a + (Sx · i), b + (Sy · j)) -FY (a + (Sx · 〇- \, b + (Sy · j)) + FY (a + (Sx · i), b + (Sy · j)) -FY (a + (Sx i) + \, b + (Sy · J))} negative (5) 10 This paper scale is applicable to China National Standards (CNS) of 8 counties (2 丨 0X297).-: ---: ------ 1 (Please read the back first (Notes for filling in this page)

, 1T

C. Printed by the Consumer Cooperative of the Central Bureau of Probability of the Ministry of Economic Affairs A7 ---------- B7 _ V. Description of the invention ([ο) Since ^^ in (4) is a constant, ) Is ignored to reduce the division operation. At the same time, the absolute 値 operation of ΣΣ |. ·. | In (4) is also equivalent to the positive part ΣΣ {.. ·} in (5) minus the negative part ΣΣ {...} „_ „, In order to reduce the absolute 値 calculation process, the other symbols are defined as in (4). Therefore, the sample subtractor 621 in the second figure performs the subtraction operation of the sampled pixel and its neighboring pixels in formula (5), and sends the result to the positive and negative number discriminator of 622. The parts are accumulated by the accumulator 623, and the negative part is subtracted by the subtractor 624. The equivalent absolute unitary operation is used to obtain a result equivalent to the expression (5). The size of the sampling interval in the simultaneous sampling subtracter can be weighed by the amount of calculation to be reduced and the accuracy of the discrimination; the larger the sampling interval, the lower the accuracy of the discrimination, but the smaller the amount of calculation; otherwise, the smaller the sampling interval , The higher the accuracy of the discrimination, but the higher the amount of calculation. In formula (5), a, b, c, and d 値 are to avoid the start and end sampling points on the edge of the image, because the information in the edge part of the general image is less important, and the quantization table determiner 625 contains Some quantization tables determined in advance to match different image complexity are based on the results of the SAD calculation by the sample subtracter 621, the accumulator 623, and the subtractor 624, and these quantization tables are selected for use by the quantizer 612. The quantization distance in the image must be determined in advance for the images of various complexity, weighing the compression ratio, and the quality of the reconstructed image. In addition, the data quantity checker 630 in the second figure is used to ensure that the compression ratio is controlled within a specified range, and it includes a data size calculator 631 11 This paper size applies to the Chinese National Standard (CNS) A4 specification (210X 297 Mm) —; --- τ ---------- Order .------ 'ηΊ Read the notes on the back before filling out this page) Printed by the Shellfish Consumer Cooperative of the Central Standards Bureau of the Ministry of Economic Affairs System A7 B7 V. Description of the invention (/) and a high-frequency coefficient deleter 632. The data size calculator 631 is used to calculate whether the amount of data encoded by the encoder exceeds the size of the predetermined storage space.疋 size, the data is directly stored in the memory 700, otherwise, the high-frequency coefficient deleter 632 directly deletes some high-frequency non-zero coefficients that are less sensitive to the human eye, causing the entropy encoder 613 to recompress Data, 2 until the size of the compressed data can be stored in the predetermined storage space. The ninth to sixteenth images further illustrate the feasible I1 generation of SAD. Among them, the ninth image is the original image and the image size is , The eleventh figure is its 3D brightness signal Relative distribution diagram, and the small diagram at the bottom left of the tenth diagram is the formula (5), where a = 6, b = 3, c = 4, and d = 9,

Sx = 9 ′ sy = 9 The shape of the sampan sampled on the right is the enlarged result of the small image, which is still close to the original ninth image to some extent. At the same time, the twelfth picture is the distribution map of the 3D brightness signal of the tenth picture. Obviously, it is still close to the style of the eleventh picture. The thirteenth to sixteenth graphs also have similar results. 'The sampled graphs are shown in the fourteenth graph, and the sampled luminance signal is shown in the tenth π graph. Fifteen figures are similar. Therefore, in order to speed up the processing speed, it is feasible to reduce the processing steps by sampling. At the same time, compared with the previous methods, the architecture of this case fully complies with the most popular baseline JPEG standard, and the amount of calculation is greatly reduced, so the compression processing time will be relatively reduced. This reduction in processing time is also very helpful for the use of digital cameras to shoot fast and continuous images. In addition to being applicable to digital cameras, this case can also be beneficial to other equipment that stores images, and the size of the sampling interval can be changed as needed. When _— ___ 12 paper exposures (210χ29 · ϋ)

A7 B7 V. Description of the invention (f >) Of course, it is also possible to perform complexity detection in both horizontal and vertical directions to improve the accuracy of judgment. This case may be modified by anyone who is familiar with the art, but it is not as bad as the protection of the patent scope attached. (Please read the precautions on the back before filling out this page} Printed by Shellfish Consumer Cooperative, Central Bureau of Standards, Ministry of Economic Affairs 13 This paper size applies to Chinese National Standard (CNS) A4 (2 丨 〇 X 297 mm)

Claims (1)

A8 B8 CS D8 Printing by the Central Consumers' Bureau of the Ministry of Economic Affairs, Consumer Cooperatives 6. Application for a patent range 1. An image compression system with a selectable compression ratio, which includes: a blanket discriminator, which inputs an image data and outputs A quantization table, and sampling analysis based on the image data to determine its complexity and the quantization table;-the encoder 'input is the image data and is electrically connected to the complexity judgment to compress the image data according to the quantization table And output an image compression data; and, a data quantity detector, which is electrically connected to the encoder, for clamping the size of the compressed image data output by the encoder, so that Not more than a predetermined range. 2. The image compression system with a selectable compression ratio as described in item 1 of the patent application park, wherein the complexity determination JI system includes: a sample subtracter, which is respectively horizontal and vertical in the image data Take one pixel every Sx and Sy pixels to calculate the absolute difference 取样 (SAD) of the pixel; — a positive and negative discriminator, which is electrically connected to the sampling subtracter to determine the positive and negative of the absolute difference 取样 of the pixel.値; an accumulator, which is electrically connected to the positive and negative number discriminator, to accumulate the absolute difference between the sampling absolute value of the pixel; and '~ a subtractor, which is electrically connected to the positive and negative discriminator, to accumulate the pixel's The absolute difference between the absolute difference of the samples and the absolute value of the absolute difference; and, a quantization table determiner is electrically connected to the subtractor, and the result of the calculated absolute difference of the samples is used as the basis for the ¥ flower table selection. A selected quantization table is output for use by the encoder. This paper size applies to China National Fengfeng (CNS > M size (210X297mm) (please read the note on the back ^ a ^ item before filling out this page); Pack. Order ABCD VI. Patent Application Scope 3. If applying for a patent An image compression system with a selectable compression ratio as described in the second item of the scope, wherein the horizontal and vertical sampling intervals Sx and Sy are between 2 and 16. One. 4. The possible application as described in the first item of the patent application scope An image compression system that selects a compression ratio, wherein the data volume detector includes: a data size calculator to calculate the female student who compresses the shadow silly compression data. When calculating that the size of the image compression data does not exceed a limit, Output the image compression data; and a high frequency coefficient deleter, which is electrically connected to the data size calculator to delete the high frequency portion of the image compression data when the size of the image compression data exceeds the limit So that the size of the image data does not exceed the limit. 5. The image compression system with a selectable compression ratio as described in item 1 of the scope of patent application, where the encoder is a standard JPEG encoder. (Please read first Note on the back, please fill out this page again)) Printed by the Shell Standard Consumer Cooperative of the Central Bureau of Standards of the Ministry of Economic Affairs 15 The paper size applies to the Chinese National Standard (CNS) A4 (210X297 mm)