CN1799492A - Quasi-lossless image compression and decompression method of wireless endoscope system - Google Patents

Abstract

The invention belongs to the technical field of medical digital image compression, and is characterized in that it is an image compression and decompression method based on first shrinkage and then color interpolation. To control the percentage of the filtered pixels in the entire image to improve the image quality, use the region of interest to determine the pixel point area that does not need to be compressed and directly enter the encoding to ensure the image quality of the region of interest. Lossless compression adopts JPEG-LS lossless compression coding. A hardware block diagram is provided accordingly. In the compression test of 7 natural images in the standard image database, it can realize the continuous adjustment of the signal-to-noise ratio from 46.37dB to infinity, and the corresponding compression rate ranges from 3.3 bits/pixel to 6.9 bits/pixel Change; when compressing the wireless endoscope image, the average image compression bit rate can be obtained at 2.18 bits/pixel, and the PSNR is greater than 47.57dB.

Description

Quasi-lossless image compression and decompression method for wireless endoscope system

technical field

The lossless and quasi-lossless image compression method and device of a wireless endoscope system belong to the technical field of medical digital image compression, and in particular relate to the technical field of high-quality digital image compression output from a digital image sensor with a format similar to a Bayer color image array. The definition of quasi-lossless image compression in the present invention is as follows: the peak signal-to-noise ratio (PSNR) of the compressed image is greater than 46.37dB, and the error value of any pixel before and after compression does not exceed 2.

Background technique

Color digital image sensors have been widely used in various high-end and low-end video fields. However, due to the large amount of color image data, the compression of color image data is very important. Generally, lossy compression is mainly used in low-end consumer image products; in the application field of high-end color image sensors, especially in the field of medical images, image quality is the first, so high-efficiency quasi-lossless and lossless images Compression research is particularly important in the application field of high-end color image sensors.

In the wireless endoscope system, in order to reduce the communication bandwidth and save the transmission power consumption of image data (when there is no image compression, the image transmission power consumption in the capsule accounts for 90% of the total power consumption in the capsule), a low-complexity The application of high-precision, high-quality image compression algorithms is very necessary. Figure 1 shows the block diagram of the simplified image acquisition, compression and wireless transmission system in the wireless endoscopic capsule system. In the wireless endoscope system, the endoscope image data in BAYER format collected by the color image sensor will be directly compressed before color interpolation, and then sent to the body by the wireless transceiver through channel coding. The Bayer color array format is the most commonly used format in digital image sensors, and Figure 2 is a diagram of the format. The external receiver decompresses the received image data and then performs color interpolation, and finally provides it to the doctor as the final diagnostic image. Since the capsule is powered by two silver oxide batteries, the image compression system is required to have low complexity to ensure low hardware overhead and low power consumption.

In the application of most traditional digital image sensors, the digital image output from the digital image sensor with a Bayer-like color image array is first interpolated to obtain full-color RGB data, and then the interpolated full-color data Perform compression processing, and finally store the compressed data locally, or send the compressed data through wireless or wired communication. This method can be shown in FIG. 3 . However, this traditional method (the method of color interpolation first and then compression) is to perform full-color interpolation on the original data in Bayer format output by the image sensor before image compression, which introduces new data redundancy before image compression. , which is very unfavorable for image compression. At present, people have begun to propose a new method of compressing first and then interpolating for image sensors, as shown in FIG. 4 . This compression method directly compresses the image data similar to the Bayer color format output by the image sensor, avoiding unnecessary data redundancy introduced by color interpolation, so this image compression method can improve the compression performance. However, these newly proposed methods of compression first and then interpolation are mainly proposed for lossy image compression, and are not suitable for applications requiring high-quality images, especially for medical image applications such as wireless endoscopes.

Contents of the invention

The purpose of the present invention is to provide a high-efficiency, low-complexity, lossless and quasi-lossless image compression/decompression method and device based on a Bayer-like color image array for a wireless endoscope system, the structure of which is shown in Figures 5 and 6 .

The method of the present invention is characterized in that: the method is an image compression and decompression method based on first compression and then color interpolation for a wireless endoscope system, and the method is sequentially composed of a compression method and a decompression method, in:

The quasi-lossless image compression method of the wireless endoscope system is to use a low-pass filter to separately reduce the three components R, B and G of the digital image data to the Bayer color image array data with many high-frequency components output by the image sensor. Pass filtering, and then a method of compressing the three image component data after filtering with lossless compression coding, wherein in the Bayer color image array, the image G component data is diamond-shaped, and the data volume accounts for the entire Bayer image. 1/2 of the amount of data, the R and B components are rectangular, each accounting for 1/4 of the entire Bayer image data amount, the lossless compression encoding refers to JPEG-LS compression encoding; the compression method contains the following steps in turn:

Step 11: Set the following input parameters for low-pass filtering to the JPEG-LS compression coding control unit in the wireless endoscope capsule:

The quality control factor, that is, the filtered pixel points account for the percentage of the entire image pixel point, expressed by q; in the low-pass filter in the wireless endoscope capsule, the columns that should be filtered are evenly selected during horizontal filtering; vertical filtering When , evenly select the rows that should be filtered;

The region of interest is represented by ROI, which contains the position and shape of the pixels for direct lossless compression without filtering, and the ROI represents the processing capability of lossless compression;

Step 12: After receiving q and ROI, the JPEG-LS compression encoding control unit in step 11 sends the data to the low-pass filter controller located in the low-pass filter in the wireless endoscope capsule and stores it;

Step 13: The foremost digital image sensor in the wireless endoscope capsule collects Bayer array color data, and sends the Bayer array color data to the In the low-pass filter described above;

Step 14: the low-pass filter control unit in the low-pass filter sends the data that does not need to be filtered to the built-in cache SRAM in the low-pass filter according to the ROI set; The G component performs a rhombus-to-rectangle conversion operation, that is, the empty points of the rhombus G component in the color image data in the Bayer format are removed, and then the remaining data are directly combined into a rectangle, so the same set of filters is used according to After the following steps use the G component low-pass filter and the B, R component filter to filter the G, B and R components respectively, store them in the built-in cache SRAM of the corresponding low-pass filter:

表示向下取整操作，即

Step 1: horizontal filtering, to weaken the high-frequency components in the horizontal direction in the original image, the horizontal filtering is performed sequentially from left to right in columns,

Indicates the rounding down operation, that is,

The image data of the first column after filtering = the original image data of the first column,

· · · ·

表示向上取整，即

Step II: perform longitudinal filtering on the new data obtained in the first step to smooth the vertical high-frequency components, and the longitudinal filtering is performed in the following order from top to bottom according to the row,

Indicates rounding up, that is,

Image data of the first row after filtering = original image data of the first row and column,

· · · ·

Step 15: The JPEG-LS compression encoding control unit sends the filtered image data stored in the SRAM built in the low-pass filter to the SRAM located in the wireless endoscope capsule to be filtered for all image data Finally, the filtered image data is sent to a JPEG-LS encoder built in the wireless endoscope capsule;

Step 16: The JPEG-LS encoder performs lossless compression on all received filtered image data according to JPEG-LS lossless compression encoding under the control of the JPEG-LS compression encoding control unit.

Step 17: After the compression operation described in step 16 is completed, the JPEG-LS encoder stores the compressed image data in the SRAM and sends it to the channel encoding unit, and outputs it to the channel located in the wireless channel after encoding. The wireless transceiver in the speculum capsule wirelessly transmits to the outside of the body;

The lossless and quasi-lossless image decompression method of the wireless endoscope system is carried out according to the following steps under the control of the JPEG-LS lossless decoding control unit outside the body:

Step 21: After the in vitro wireless transceiver device receives the filtered and compressed data from the corresponding device in the wireless endoscope capsule, the data is sent to a JPEG-LS decoder for decoding;

Step 22: After the JPEG-LS decoder decodes the compressed data in step 21, send the G component and the B and R components to the G component reconstruction filter and the B and R component reconstruction filter respectively for reconstruction filtering ;

Step 23: The reconstruction filter described in step 22 performs reconstruction filtering on the G component and the B and R components respectively according to the following steps:

Step 1: the reconstruction filter sends data that does not need reconstruction filtering in the set ROI region to the built-in cache SRAM in the reconstruction filter; for data that needs reconstruction filtering, according to the set q The value is followed by the following vertical reconstruction filtering and horizontal reconstruction filtering;

Step II: vertical reconstruction filtering: carry out sequentially from bottom to top according to the following formula, and restore the vertical original data;

Image data value of line m after reconstruction=2*original image data value of line m-original image data value of line m-1, image data value of line m-1 after reconstruction=2*original line m-1 Image data value - original m-2 line image data value,

· · · ·

Image data value of the second row after reconstruction=2*original image data value of the second row-original image data value of the first row,

Image data value of the first line after reconstruction = original image data value of the first line,

The third step: horizontal reconstruction filtering: the new data obtained in the second step is carried out from right to left in sequence according to the following formula, to complete the recovery of the horizontal direction data;

Image data value of column m after reconstruction = 2* original image data value of column m - original image data value of column m-1, image data value of column m-1 after reconstruction = 2* original column m-1 Image data value - the image data value of the original m-2th column,

· · · ·

Image data value of the second column after reconstruction = 2* original image data value of the second column - original image data value of the first column,

Image data value of column 1 after reconstruction = original image data value of column 1;

Step IV: send the image data after reconstruction filtering to the built-in SRAM;

Step 24: the reconstruction filter sends the G component data after reconstruction filtering to a rhombus-to-rectangular converter for transformation;

Step 25: Send the reconstructed and filtered B and R component data obtained in step 23 and the transformed G component data obtained in step 24 to an adder for addition to obtain the restored original Bayer color image data;

Step 26: send the data obtained in step 25 to a color interpolation processor to interpolate according to the Laplace method to obtain full-color image data.

Described lossless compression encoding can be JPEG-LS, JPEG2000 lossless compression part and FELICS algorithm (P.G.Howard and J.S.Vitter, Fast and Efficient Lossless Image Compression [A], IEEE DataCompression Conference [C], USA, 1993: 351-360 .) in any one.

The present invention is realized with the following devices: said device is a wireless endoscope capsule located in the body, which includes:

The internal ASIC chip contains: rhombus-to-rectangular transformation unit, G component image data low-pass filter unit, B, R component low-pass filter unit, JPEG-LS lossless compression unit and JPEG-LS lossless compression control unit, of which:

A rhombus-to-rectangular conversion unit, the input of which is connected to an image sensor that can output Bayer color image data, to remove the blank points in the G component data of the rhombus, and then combine the remaining data directly into a rectangle before output;

A low-pass filter, including a G component and a low-pass filter unit of B and R components, the low-pass filter includes: a low-pass filter controller, a row counter, a column counter, an input synchronous extraction circuit, and a buffer for two rows of image data SRAM, G, B, R component registers, data selector A, data selector B and filter operator, where:

A low-pass filter controller is provided with: a quality control factor q value input terminal, said q refers to: the filtered pixel points account for the percentage of the entire image pixel point; an ROI value input terminal of a region of interest, said ROI value refers to Contains the position and shape of pixels without filtering for direct lossless compression;

A row counter and a column counter, the output terminals of which are respectively connected to the corresponding input terminals of the low-pass filter controller;

The input synchronous extractor is provided with: a Bayer image data input terminal, which is connected to the corresponding output terminal of the image sensor unit; a synchronous signal input terminal including a line/field synchronous signal and a synchronous clock of each pixel point, which is connected to the synchronous signal input terminal. The input terminal is connected with the corresponding output terminal of the image sensor unit; the input synchronous extractor sends the extracted corresponding row and column information to the row and column counter unit respectively; the input synchronous extractor is controlled row by row under the control of the synchronous signal Or output the 8-bit Bayer image data of the current row and column column by column;

The cache SRAM is provided with an address signal input terminal, which is connected to the corresponding output terminal of the low-pass filter controller, and is also provided with a data input terminal and an output terminal of the previous row or column;

The G component register, the B component register and the R component register are each provided with a strobe signal input terminal, and the input terminal is connected to the corresponding output terminal of the low-pass filter controller; the data input terminal of the previous row or column is provided, and the The input end is connected to the corresponding output end of the buffer SRAM; a data input end of the current row or column is also provided, and the input end is connected to the corresponding output end of the input synchronous extraction unit;

The data selector A is provided with an input terminal corresponding to the output data of the current row or column and the previous row or column of the G, B, R component registers;

The data input terminal of the arithmetic unit is connected to the corresponding data output terminal of the data selector A, and the image data of each input component is sequentially filtered horizontally and vertically according to the following operation method:

Step 1: horizontal filtering, to weaken the high-frequency components in the horizontal direction in the original image, the horizontal filtering is performed sequentially from left to right in columns, Indicates the rounding down operation, that is,

The image data of the first column after filtering = the original image data of the first column,

· · · ·

Step II: perform longitudinal filtering on the new data obtained in the first step to smooth the vertical high-frequency components, and the longitudinal filtering is performed in the following order from top to bottom according to the row, Indicates rounding up, that is,

Image data of the first row after filtering = original image data of the first row and column,

· · · ·

Data selector B is provided with a strobe signal input terminal, which is connected to the corresponding output terminal of the low-pass filter controller, so as to input the input of image data that does not need to be filtered and has been filtered; A data input terminal, which is connected to the Bayer color image data output terminal of the current row or column of the input synchronous extractor, and is also provided with a filtering data input terminal, which is connected to the corresponding data output terminal of the arithmetic unit connected; in addition, there is a filtered image data output terminal of the previous row or column, which is connected to the image data input terminal of the previous row or column of the cache SRAM, and the image data of the previous row or column The output terminal is also a Bayer color image data output terminal processed by the low-pass filter;

SRAM is provided with a Bayer color image data input terminal, which is connected to the corresponding data output terminal of the low-pass filter; a control signal input terminal is also provided, and the input terminal is connected to the control of the JPEG-LS lossless encoding The corresponding output terminals of the unit are connected;

JPEG-LS lossless compression encoder, encoded according to the JEPG-LS lossless compression standard file, the encoder contains:

The cache memory of four parameters A, B, C, N, each input end is connected with the corresponding output end of the four parameters A, B, C, N set in the SRAM, and the four parameters are all connected to a Corresponding to the context conditions, A refers to the cumulative absolute error value under the current context conditions, C refers to the average error value, N refers to the total number of occurrences of the current context, and the B parameter is an intermediate quantity introduced to simplify the calculation, A, B, C , The N parameter is used for self-adaptive adjustment of the image's own content, and is used to reduce the error of the predicted value;

A run-length scanning and run-length encoding circuit, the run-length mode setting end of the circuit is connected to the output end of the run-length mode value of the SRAM cache;

A context decision circuit, which is interconnected with the control unit of the JPEG-LS lossless compression coding, and has an input end connected with the data output end of the SRAM. The decision circuit calculates and quantizes the local gradient according to the context content of the compressed point, as well as the fusion of the quantized gradient and the selection of the compression mode; the local gradient refers to the three gradients formed by four points around the pixel to be compressed Value (that is, the difference between two pixels), the local gradient contains three gradient components, which correspond to two horizontal gradients of the upper right and upper, upper left and upper left of the pixel to be compressed, and a vertical gradient of upper left and upper left of the pixel to be compressed;

An error prediction circuit, the control signal data end of the circuit is connected to the corresponding output end of the control unit of the JPEG-LS lossless compression encoding, and the input end is connected to the data output end of the SRAM, and the other output port is connected to the parameter calculation unit, To read and update the A, B, C, N parameters, the error prediction circuit performs rough prediction according to the local gradient and the values of the surrounding four points, and according to the context value of the current pixel corresponding to A, B, C, N The parameters fine-tune the initial value of the forecast to finally determine the forecast value, and then update the A, B, C, and N parameters;

The calculation unit of parameters A, B, C and N: there are four data ports, and the ports are connected to the corresponding output terminals of the four parameters A, B, C, N buffer units; an error input port is provided , the port is connected to the error prediction circuit; this unit completes the operation of the error prediction value input from the error circuit and the corresponding value read from the parameter A, B, C, N buffer area unit, and sends the result obtained by the operation to Back to parameter A, B, C, N buffer unit and error prediction unit storage;

The control unit of JEPG-LS lossless compression coding is provided with the output end of described q, ROI value, and this output end is connected with the corresponding output end of the low-pass filter controller in the described low-pass filter; Meanwhile, this JPEG-LS The control unit of the LS lossless compression coding is connected with the low-pass filter, the SRAM and the context decision circuit, the error control circuit and the run-length scanning and coding circuit in the JPEG-LS compression coder.

Invention effect is as follows:

1) This compression method provides an efficient, low-complexity, Bayer-like color image array-based lossless and quasi-lossless image compression for wireless endoscopy systems. In order to demonstrate the effectiveness of the compression method, the Bayer color format image data is partly obtained from the standard image library of 'lena', 'baboon', 'airplane', 'house', 'lake', 'peppers' and 'splash', etc. The 7 standard images are obtained by sampling operation; the other part of the images is from the 6 collected wireless endoscopic images.

Adopt this compression method to natural image in the standard image storehouse, when quality control factor q=1, the average compression code rate of color image is 3.35 bits/pixel, the peak signal-to-noise ratio (PSNR) of image＞=46.37dB; The wireless endoscope image compression method can provide an average compression code rate of 2.18 bits/pixel for the color image, and a peak signal-to-noise ratio (PSNR) of the image>=47.57dB.

2) The implementation complexity of the low-pass filter involved in this method is very low, and it can be realized only by addition and shift operations.

3) because the compression method that the present invention proposes is to compress before color interpolation, compares with the conventional method of first color interpolation and then compression, the G component data that needs to be compressed has only half of the conventional method G component data volume, and R and B component Both are only a quarter of the conventional method, so the method greatly reduces the amount of compressed data, and reduces the complexity and storage capacity of hardware implementation.

4) It can provide quasi-lossless compression or lossless compressed image data whose signal-to-noise ratio changes continuously within a certain range. The results of compressing images in the standard image library show that quasi-lossless compression and lossless compression can be obtained by controlling the quality factor of the input parameter of the low-pass filter. For image data, the corresponding compression ratio varies from an average of 3.35 bits/pixel to 6.9 bits/pixel.

5) It can provide functions such as lossless compression for the specified region of interest ROI area, that is, it supports the region of interest (ROI) processing function. Control ports by compressing the region of interest (ROI). The shape of the ROI can be arbitrary, such as: rectangle, circle, ellipse, and any shape.

In order to verify the effectiveness of the algorithm, Table 1 lists the results of comparing the proposed algorithm with the following algorithms.

a) 'compression method of the present invention (q=1), lossless compression encoding JPEG-LS': the compression method proposed by the present invention, wherein the lossless compression encoding part adopts JPEG-LS compression encoding, and the quality control factor of the low-pass filter q=100%.

b) 'compression method of the present invention (q=1), lossless compression coding JPEG2000': the compression method proposed by the present invention, wherein the lossless compression coding part adopts JPEG2000 compression coding, and the quality control factor q=100% of the low-pass filter .

c) 'compression method of the present invention (q=1), lossless compression coding FELICS': the compression method proposed by the present invention, wherein the lossless compression coding part adopts FELICS compression coding, and the quality control factor q=100% of the low-pass filter .

d) 'JPEG-LS quasi-lossless compression (near parameter is 2)': BAYER data is directly compressed by the JPEG-LS quasi-lossless compression encoder, where the near parameter is 2, which is the error value of each pixel of the compressed image Not greater than 2.

e) 'compression method of the present invention (q=0.25), lossless compression coding JPEG-LS': the compression method proposed by the present invention, wherein the lossless compression coding part adopts JPEG-LS compression coding, and the quality control factor q of the low-pass filter = 25%.

f)' Structure conversion method in reference (Chin Chye Koh, Jayanta Mukherjee and Sanjit K.Mitra, New Efficient Methods of Image Compression in Digital Cameras with Color Filter Array[J] (digital for color filter array image data output High-efficiency image compression method for cameras), IEEE Trans.ConsumerElectronics, Nov 2003, 49(4): 1448-1456.)': BAYER data is filtered by the structural conversion filter in the literature and then losslessly compressed and encoded by JPEG-LS.

g) 'JPEG-LS direct compression method': image data in Bayer color format is directly compressed using JPEG-LS.

f) 'JPEG2000 direct compression': JPEG2000 is used to directly compress image data in Bayer color format.

Table 1 is the result of compressing 7 standard images in the image bank, and it shows: the average compression code rate of the compression method (q=1) proposed by the present invention reaches 3.35 bits/pixel, and the average PSNR value reaches 46.43dB, And the error of each pixel point value after compression can not be greater than 2, so this algorithm has guaranteed the quality of compressed image; The compression code rate (q=1) that the compression method that the present invention proposes obtains will be lower than adopting JPEG-LS and JPEG-LS and JPEG2000 lossless coder is to the compression code rate under the direct compression situation of Bayer color format image data, and image still keeps very high signal-to-noise ratio (greater than 46.37dB) after restoring; Lossless compression coding in the compression method adopts JPEG-LS to have better compression performance than adopting JPEG2000 and FELICS; When q=0.25, the compression performance of the compression method proposed by the present invention will be higher than "structure conversion" in the reference Algorithm; when q=1, the compression performance of the compression method proposed by the present invention is higher than that of the quasi-lossless compression of JPEG-LS (near parameter is 2).

Table 2 is the experimental results of compressing six typical wireless endoscope images. The results also show that the compression algorithm proposed by the present invention has higher compression performance than the JPEG-LS quasi-lossless compression (near parameter is 2) and the "structure conversion" algorithm in the reference.

Experimental results show that from the perspective of compression performance and hardware implementation complexity, the compression method proposed by the present invention is the most suitable algorithm for directly performing quasi-lossless compression on Bayer color array image data, and is therefore suitable for wireless endoscope capsule systems for low Complexity, low power requirements.

Table 1. Experimental results and comparison of 7 images compressed in the standard image library Image (512*512) airplane baboon house the lake lena peppers splash Compression method (q=1) of the present invention, lossless compression coding JPEG-LS PSNR(dB) 46.4160 46.3914 46.4328 46.4011 46.4099 46.5528 46.5374 CR(bpp) 2.8963 4.6882 3.3301 3.7741 3.1440 3.2130 2.4590 Compression method (q=1) of the present invention, lossless compression coding JPEG2000 PSNR(dB) 46.4160 46.3914 46.4328 46.4011 46.4099 46.5528 46.5374 CR(bpp) 4.7924 6.6782 5.0576 5.7585 5.0936 5.1689 4.2871 Compression method (q=1) of the present invention, lossless compression coding FELICS PSNR(db) 46.4160 46.3914 46.4328 46.4011 46.4099 46.5528 46.5374 CR(bpp) 4.1916 5.6328 5.4711 5.5309 6.5132 7.0018 6.8462 JPEG-LS quasi-lossless compression equal to 2 PSNR(dB) 45.1718 45.3809 45.1541 45.1337 45.1239 45.2419 45.2670 CR(bpp) 3.0123 6.6121 3.9325 4.4080 4.8955 5.2372 4.8723 Compression method (q=0.25) of the present invention, lossless compression coding JPEG-LS PSNR(db) 52.4090 52.3816 52.4528 52.3953 52.4134 525909 52.6665 CR(bpp) 4.3952 6.3473 5.2241 5.6612 4.7357 4.7772 3.7493 Wrong reference! Reference source not found. Struct transformation methods in PSNR(db) 52.4083 52.3809 52.3770 52.3889 52.3387 525615 52.5792 CR(bpp) 4.5939 6.6935 5.8489 5.8734 5.1658 4.9416 4.1000 JPEG-LS direct compression PSNR(dB) ∞ ∞ ∞ ∞ ∞ ∞ ∞ CR(bpp) 5.2069 7.5596 6.2539 6.8519 7.4460 7.6901 7.4404 JPEG2000 direct compression PSNR(dB) ∞ ∞ ∞ ∞ ∞ ∞ ∞ CR(bpp) 4.5739 5.7557 5.2302 5.8539 5.4564 5.6753 5.0229

CR means compression code rate, ∝ means infinity

Table 2 Comparison of several image compression algorithms (6 typical wireless endoscopic images) Image (256×256) (a) (b) (c) (d) (e) (f) Compression method (q=1) of the present invention, lossless compression coding JPEG-LS PSNR(db)CR(bit/pixel) 46.8911.962 46.8822.176 46.9212.223 46.8752.256 46.9042.081 46.9132.411 Compression method (q=0.25) of the present invention, lossless compression coding JPEG-LS PSNR(db)CR(bit/pixel) 52.8153.095 52.7983.331 53.0063.581 52.9623.636 52.9913.209 53.0173.752 Quasi-lossless compression of JPEG-LS (near parameter is 2) PSNR(db)CR(bits/pixel) 45.3872.168 45.7342.412 45.7362.228 45.2192.6311 45.5812.174 45.5922.415 Structural Transformation Methods in References PSNR(db)CR(bits/pixel) 51.7313.815 51.6984.022 51.8124.206 51.2964.318 51.7994.031 51.8254.527

In addition to low complexity and high compression performance, the compression method proposed by the invention also has the functions of adjustable compressed image quality and lossless compression of ROI. Figure 15 shows the relationship between the adjustable image quality PSNR and the compression bit rate of the three images (baboon, lena and airplane), the image quality PSNR is continuously adjustable from 46.37dB to infinity, and the corresponding average compression bit rate is from 3.35 bits /pixel to 6.9 bits/pixel range change.

Description of drawings

Fig. 1 Simplified image compression system model in wireless endoscope;

Figure 2 BAYER color array format diagram;

Fig. 3 traditional image acquisition, compression and transmission system based on digital image sensor (compression method after first color interpolation);

Fig. 4 new image acquisition, compression and transmission system based on digital image sensor (color interpolation method after compression first);

Fig. 5 is a block diagram of a compression structure based on a color interpolation method after compression for a wireless endoscope system proposed by the present invention;

Fig. 6 is a block diagram of decompression structure based on first compression and then color interpolation method for wireless endoscope system proposed by the present invention;

The transformation diagram of the G component of the image data in Bayer color format from rhombus to rectangle in Fig. 7;

Figure 8 is based on a digital image lossless and quasi-lossless image compression/decompression method and device example diagram of a similar Bayer color image array;

Figure 9a Transverse filtering;

Figure 9b longitudinal filtering;

Figure 9c longitudinal reconstruction filtering;

Figure 9d horizontal reconstruction filtering;

When Fig. 10 (a) q=50%, the number of filtered rows selected by horizontal filter; (b) when q=25%, the number of columns filtered by vertical filter;

Algorithm flow chart during image quality adjustment in Fig. 11;

Figure 12 is the selected ROI in the dotted box;

The VLSI structure of Fig. 13 image compression method;

The hardware realization structure of low-pass filter in Fig. 14 image compression algorithm;

The relationship between the compression code rate and the compression ratio when Fig. 15 adjusts the quality control factor of image compression;

The program flowchart of the compression method that Fig. 16 present invention proposes;

The program flowchart of the decompression method proposed by the present invention in Fig. 17;

Detailed ways:

Fig. 1 is a block diagram of the simplified image acquisition, compression and wireless transmission system of the wireless endoscopic capsule system. The Bayer color format image data output by the image sensor in the wireless endoscope capsule is directly compressed, and then sent to the channel coding unit, after channel coding, it is output to the wireless transceiver, and sent to the body in the form of radio waves. The capsule can also receive control command data from the outside through the wireless transceiver, and then send it to the channel decoding unit, and the decoded data is output to the control unit, and the control unit controls the next step of the circuit in the capsule according to the received control command.

Figure 2 is a BAYER format color image array output by the image sensor. Among them, the G component data of the image is diamond-shaped, and the data volume accounts for 1/2 of the entire Bayer image data volume, and the R and B components are rectangular, each accounting for 1/4 of the entire Bayer image data volume.

Figure 3 is the traditional method of color interpolation first and then compression. The color image data output by the image sensor in Bayer format is first processed by color interpolation to form full-color RGB image data, and then the full-color image data is compressed, and the compressed data is stored locally or through wired or wireless communication. Send it out, and finally decompress the compressed data stored locally or obtained through wired or wireless means to get the restored full-color image data.

Figure 4. New compression method based on color interpolation after compression. The Bayer color format image data output by the image sensor is directly compressed, the compressed data will be stored locally or sent out through wired or wireless communication, and finally the compressed data stored locally or obtained through wired or wireless reception is decompressed Compress, and finally perform color interpolation processing on the decompressed Bayer color format image data to form full-color RGB image data, and restore it to full-color RGB image data.

Fig. 5 is a compression structure based on first compression and then color interpolation for the wireless endoscope system proposed by the present invention. The image sensor outputs color image data in Bayer format. The G component in the data is first converted from a rhombus to a rectangle, then sent to a low-pass filter for smoothing, and finally sent to a lossless compression module for lossless compression, where R and B The components are originally rectangular, so they will be low-pass filtered directly through the low-pass filter, and finally sent to the lossless compression module for lossless compression. The smoothing filter has two control parameters: quality control factor and region of interest (ROI) parameter.

In Figure 6, the compressed data output in Figure 5 is decompressed, and then the G component of the decompressed output is transformed from a rectangle to a rhombus, and then sent to the reconstruction filter for reconstruction, while the R and B components of the decompressed output The components are directly sent to the reconstruction filter for reconstruction. The reconstruction filter has two input control parameters that are the same as the low-pass filter of the compression part: the quality control factor and the region of interest (ROI) parameter. For the two The G, R and B components output by the reconstruction filter are recombined by the adder into Bayer format color image data and output to the color interpolation processing unit for color interpolation, and finally obtain full-color RGB data output.

Figure 7 removes the empty points of the rhombus G component in the color image data in Bayer format, and then directly combines the remaining data points into a rectangle.

FIG. 8 is a specific implementation device of the high-efficiency, low-complexity, lossless and quasi-lossless image compression/decompression method based on a Bayer-like color image array for a wireless endoscope system according to the present invention. The wireless endoscope capsule includes data acquisition, the compression device proposed by the present invention (including a rhombus-to-rectangular transformation unit, a low-pass filter, and a JPEG-LS encoder) and a wireless transmitting and receiving device A; the wireless capsule includes the present invention The proposed data decompression and recovery device, and the wireless receiving and transmitting device B group. The data collection and compression device in the capsule mainly completes the process of collecting and compressing image data in Bayer color format. The front end of the device is a CCD/CMOS digital image sensor, which outputs image data in Bayer color format after image collection. Because the Bayer color format data has many high-frequency components, the unprocessed original data is not suitable for direct compression, so the original data needs to be processed by low-pass smoothing filter. The pixel points of the G component in the Bayer color format image data are arranged in a diamond shape. Before filtering the G component, it needs to be transformed from a diamond shape to a rectangle, and then perform low-pass filtering. The R and B components can be directly low-pass filtered, and finally filtered. The last three components will be respectively sent to the lossless compression unit for compression, and the compression unit adopts JPEG-LS lossless compression coding. The compression device is implemented using an application-specific integrated circuit (ASIC). The compressed data will be transmitted out of the capsule in the form of radio waves through the wireless receiving and transmitting device A.

The wireless transmitting and receiving device B outside the capsule receives the compressed image data transmitted from the capsule, and the data is decompressed by the decompression method proposed by the present invention, and the decompressed data is finally restored to full-color RGB image data through color interpolation processing. The decompression means includes a reconstruction filter, a diamond-to-rectangular transformation unit, an adder, and a JPEG-LS decoder. The decompression part is implemented by software in a specific example.

First define an objective comparison standard for image compression, see the following formula (1):

$\mathrm{<span\; class="notranslate">\; PSNR</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 10</span>}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; 255</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}}\underset{\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; W</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}\underset{\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; h</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; I</span>}}_{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

Low pass filter design:

代表第m列滤波后数据， 代表第m列未滤波原始数据，在横向滤波过程中，除第一列数据外的其它数据都要经过滤波处理，具体操作可由公式(2)来表达，其中

表示取整操作，有 Firstly, the G component is transformed from a rhombus to a rectangle as shown in Figure 7, and then G, B, and R are filtered separately, and the same set of filters is used. The set of filters is divided into two parts: horizontal filtering see Fig. 9a and vertical filtering see Fig. 9b. The figure shows 4×4 bayer data. In Figure 9a, '●' represents the original data in the first column,

Represents the mth column filtered data, Represents the unfiltered original data in the mth column. In the process of horizontal filtering, all data except the first column of data must be filtered. The specific operation can be expressed by formula (2), where

Indicates the rounding operation, there is

The order of filtering operations is from left to right.

The image data of the first column after filtering = the original image data of the first column,

· · · ·

代表第m行滤波后数据，

代表第m行未滤波原始数据，具体操作可用公式(3)表示。其中 是取整操作，但 滤波过程按照从上到下的顺序。After horizontal filtering, the high-frequency components in the horizontal direction in the original image are greatly reduced. In order to further smooth the high-frequency components in the vertical direction, the filtered output data is connected to a vertical filter. The filtering process is shown in Figure 9b. In Figure 9b, '●' represents the first row of original data,

Represents the filtered data of the mth row,

Represents the unfiltered raw data of the mth row, and the specific operation can be expressed by formula (3). in is a rounding operation, but The filtering process follows the order from top to bottom.

Image data of the first row after filtering = original image data of the first row and column

· · · ·

Reconstruction filter design:

The reconstruction filter bank also includes two filters: a vertical reconstruction filter and a horizontal reconstruction filter. These two parts of the reconstruction filter correspond to the previous vertical filtering and horizontal filtering respectively. The purpose of vertical reconstruction filtering is to restore the original vertical data. The filtering operation process is shown in Figure 9c. The reconstruction process is carried out row by row, and the data values of the previous row are restored one by one in the order from bottom to top until the first row. The specific filtering process can be expressed by formula 4 . During vertical reconstruction, the serial number in the formula represents the serial number of the row.

Image data value of line m after reconstruction=2*original image data value of line m-original image data value of line m-1

Image data value of line m-1 after reconstruction=2*original image data value of line m-1-original image data value of line m-2

· · · ·

Image data value of the second row after reconstruction = 2* original image data value of the second row - original image data value of the first row

Image data value of line 1 after reconstruction = original image data value of line 1 (4)

The horizontal reconstruction filter completes the function of data recovery in the horizontal direction. The reconstruction process is carried out by row, and the data in the previous column is restored one by one until the first column in the order from right to left. The process can also be described by formula (4), but in the formula The serial number of represents the serial number of the column, and its filtering operation process is shown in Figure 9d.

Error Analysis:

The source of the error in the proposed algorithm is the horizontal and vertical filtering operations in the low-pass filter, that is, the rounding operations in formulas (2) and (3). The precision of 8 bits/pixel is adopted in the actual wireless endoscope system. The error e _H caused by the horizontal filtering formula (2) can be expressed by the formula (5). In the formula, x represents the sum of the values of two adjacent pixels during the filtering process, and is an 8-bit integer. It can be known from this formula: e _H =0 or 1.

Similarly, the error e _V caused by the longitudinal filtering formula (3) can be expressed by formula (6). It can be known from this formula: e _V =0 or -1.

Therefore, after horizontal and vertical filtering, the total error can be described by the following formula (7). where in the formula

部分是由于纵向滤波引入误差再通过横向重构滤波器扩大后的最终误差，而

则是由于横向滤波引入的误差。因此通过公式(6)可知低通滤波器总的引入误差将不会大于2。

Partly due to the error introduced by the vertical filter and then the final error enlarged by the horizontal reconstruction filter, while

It is the error introduced by transverse filtering. Therefore, it can be seen from formula (6) that the total error introduced by the low-pass filter will not be greater than 2.

Through the statistical results of a large number of images, the probability distribution of the errors introduced by the horizontal filter and the vertical filter as shown in the formula (8) can be obtained.

${\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; v</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 8</span>}<span\; class="notranslate">\; )</span>$

Combining the error formula (7) of the low-pass filter and the probability distribution of errors introduced by horizontal and vertical filtering, it can be obtained that the probability distribution of the error value introduced by the low-pass filter is shown in the following formula (9).

$\left\{\begin{array}{c}{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\\ {\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span>\\ <span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 2</span>}}\\ {\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 0</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; V</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; p</span>}}_{\mathrm{<span\; class="notranslate">\; h</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; 4</span>}}\end{array}\right.<span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 9</span>}<span\; class="notranslate">\; )</span>$

Therefore, according to the error probability distribution formula (8) of the low-pass filter and the definition of PSNR (1), it can be concluded that the theoretical PSNR value of the decompressed reconstructed image data will be 46.37dB, see formula (10). Therefore, the quasi-lossless compression algorithm proposed in this paper theoretically not only ensures that the error value of each pixel after decompression and reconstruction does not exceed 2, but also ensures that the PSNR of the decompression and reconstruction image is greater than 46.37dB.

$\mathrm{<span\; class="notranslate">\; PSNR</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 10</span>}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; 255</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}}<span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; 10</span>}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; 255</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 46.37</span>}\mathrm{<span\; class="notranslate">\; dB</span>}<span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 10</span>}<span\; class="notranslate">\; )</span>$

In the algorithm, the quality factor of the filter can be controlled to select the number of pixels to be filtered to adjust the quality and compression ratio of the compressed image. The percentage of filtered pixels in the algorithm to the entire image pixels will be used as the input parameter of the low-pass filter, that is, the quality control factor. If the quality factor is introduced into formula (1), formula (11) for adjustable image quality can be obtained. In the formula, q represents the quality factor, and q≤1. When q=0, it means lossless compression, and PSNR tends to infinity; when q=1, it means that all pixels will be filtered, and PSNR=46.37dB at this time. Therefore, the algorithm can provide PSNR adjustment from about 46.37dB to infinity.

$\mathrm{<span\; class="notranslate">\; PSNR</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 10</span>}{\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}<span\; class="notranslate">\; (</span>\frac{{\mathrm{<span\; class="notranslate">\; 255</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}}<span\; class="notranslate">\; [</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; p</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; W</span>}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&times;</span>{\mathrm{<span\; class="notranslate">\; 2</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; ]</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 46.37</span>}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; 10</span>}\mathrm{<span\; class="notranslate">\; log</span>}}_{\mathrm{<span\; class="notranslate">\; 10</span>}}\mathrm{<span\; class="notranslate">\; q</span>}<span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; .</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

In order to simplify the selection of the filtered points in practical applications, the present invention provides a simple method to select the filtered points. Adjust the quality of the compressed image by changing the selection of the filter point to the selection of the filter row or column. During horizontal filtering, select the filtered columns evenly according to the quality factor; during vertical filtering, select the filtered rows as uniformly as possible. Figure 10 shows the number of filtered rows and columns selected by the vertical and horizontal filters when q=25%. The algorithm process when performing image quality adjustment can be seen in the flow chart in Figure 11.

In order to ensure the quality of ROI image data, the algorithm provides the ability to compress ROI losslessly. According to the ROI position and shape information contained in the input ROI parameters of the filter, the algorithm does not filter the points in the ROI to achieve lossless compression of the ROI. Figure 12 shows an example of lossless compression for a ROI shown in a 2×2 virtual frame, where the four points G44, R45, B54 and G55 will not be filtered.

The VLSI structure of the image compression circuit is shown in Figure 13. The Bayer array color image data output from the image sensor and various synchronous signals will firstly be low-pass smoothed by a low-pass filter on the G, B and R components, and then stored in the SRAM until all pixels are filtered. Then the JPEG-LS module reads the filtered data from the SRAM for lossless compression, and the compressed data will be stored back in the SRAM again, so that the wireless transmitting and receiving device can transmit to the outside of the capsule at different code rates. The hardware implementation of JPEG-LS mainly includes the following parts:

a) Control unit for JPEG-LS

It mainly implements the filtering and data storage control of the input Bayer image data, as well as data compression, and judges the encoding mode that the current compression point should enter based on the output of the context decision-making unit. It also controls the entire JPEG-LS compression module. clock management.

b) Context Decision Unit

This module calculates and quantizes the local gradient according to the context content of the current compressed point, and selects the fusion of the quantized gradient and the compression mode. The result of the decision will be sent back to the control unit. The local gradient refers to the The three gradient values formed by the four points around the pixel (that is, the difference between two pixels), the local gradient contains three gradient components, corresponding to the two horizontal gradients of the upper right and upper, upper left and upper left of the pixel to be compressed, and a vertical gradient from top left to right left. .

c) Error prediction

This unit first completes the median edge detection of the compressed point, that is, the fixed prediction of the compressed point value, and performs adaptive correction on the fixed predicted value, then calculates the error of the predicted value, and classifies and compares the error value. map.

d) Calculation unit for parameters A, B, C and N

In error prediction and calculation, four sets of parameters A, B, C, and N are needed. The parameter A refers to the cumulative absolute error value under the current context, C refers to the average error value, and N refers to the total error value in the current context. The number of times, the B parameter is an intermediate quantity introduced to simplify the calculation. These four groups of parameters A, B, C, and N are used for adaptive adjustment of the image according to its own content, and are used to reduce the error of the predicted value. This unit completes the The error prediction value input from the error circuit and the corresponding value read from the parameter A, B, C, N buffer area units are operated, and the results obtained by the operation are sent back to the parameter A, B, C, N buffer area units to store them. Four groups of prediction parameters are stored in the buffer area of the respective parameters and the error prediction unit, the required memory size=368×16 (parameter A)+368×6 (parameter B)+368×8 (parameter C)+368×6 (parameter N) = 13248 bits.

e) Golomb coding

In the normal coding mode of JPEG-LS, Golomb coding with a limited codeword length is performed on the prediction error value.

f) Run-length scanning and run-length encoding

Perform a run-length scan on the compression point that enters the run-length encoding mode, and perform Golomb encoding on the scanned run length. Therefore, the total memory overhead of the image compression part is 322×288×8+13248=755136 bits.

The hardware structure of the core filter part is shown in Figure 14, and the implementation is very simple. The input synchronous signal mainly includes: line/field synchronous signal, and the synchronous clock signal output by each pixel, etc. The line and field synchronous signal and the synchronous clock signal are mainly used to calculate the row and column where the current output image data is located, and provide a filter control unit to decide whether the current pixel is to be filtered by the low-pass filter, and through the selection of the data selector B, to Realize the lossless compression of the image data in the ROI. Both the filtered and unfiltered data in the ROI will be stored in SRAM. The filtering process only needs an adder with a width of 8 bits, and the input selection of the adder is realized through the data selector A, that is, the three-way image components of G, B and R are selected to be filtered respectively. The hardware overhead of the filter is very low. For each pixel to be filtered, only two 8-bit wide addition operations and one SRAM write operation are required. Each color component only needs two registers to store adjacent points. pixel value.

Claims (2)

1. The quasi-lossless image compression and decompression method of wireless endoscope system, it is characterized in that the method is a kind of image compression and decompression method based on color interpolation after first compression for wireless endoscope system, described method consists of The compression method and the decompression method are composed in sequence, among which: The quasi-lossless image compression method of the wireless endoscope system is to use a low-pass filter to separately reduce the three components R, B and G of the digital image data to the Bayer color image array data with many high-frequency components output by the image sensor. Pass filtering, and then a method of compressing the three image component data after filtering with lossless compression coding, wherein in the Bayer color image array, the image G component data is diamond-shaped, and the data volume accounts for the entire Bayer image. 1/2 of the amount of data, the R and B components are rectangular, each accounting for 1/4 of the entire Bayer image data amount, the lossless compression encoding refers to JPEG-LS compression encoding; the compression method contains the following steps in turn: Step 11: Set the following input parameters for low-pass filtering to the JPEG-LS compression coding control unit in the wireless endoscope capsule: The quality control factor, that is, the filtered pixel points account for the percentage of the entire image pixel point, expressed by q; in the low-pass filter in the wireless endoscope capsule, the columns that should be filtered are evenly selected during horizontal filtering; vertical filtering When , evenly select the rows that should be filtered; The region of interest is represented by ROI, which contains the position and shape of the pixels for direct lossless compression without filtering, and the ROI represents the processing capability of lossless compression; Step 12: After receiving q and ROI, the JPEG-LS compression encoding control unit in step 11 sends the data to the low-pass filter controller located in the low-pass filter in the wireless endoscope capsule and stores it; Step 13: The foremost digital image sensor in the wireless endoscope capsule collects Bayer array color data, and sends the Bayer array color data to the In the low-pass filter described above; Step 14: the low-pass filter control unit in the low-pass filter sends the data that does not need to be filtered to the built-in cache SRAM in the low-pass filter according to the ROI set; The G component performs a rhombus-to-rectangle conversion operation, that is, the empty points of the rhombus G component in the color image data in the Bayer format are removed, and then the remaining data are directly combined into a rectangle, so the same set of filters is used according to After the following steps use the G component low-pass filter and the B, R component filter to filter the G, B and R components respectively, store them in the built-in cache SRAM of the corresponding low-pass filter: Step 1: horizontal filtering, to weaken the high-frequency components in the horizontal direction in the original image, the horizontal filtering is performed sequentially from left to right in columns,

Indicates the rounding down operation, that is,

The image data of the first column after filtering = the original image data of the first column, Step II: perform longitudinal filtering on the new data obtained in the first step to smooth the vertical high-frequency components, and the longitudinal filtering is performed in the following order from top to bottom according to the row, Indicates rounding up, that is,

Image data of the first row after filtering = original image data of the first row and column, Step 15: The JPEG-LS compression encoding control unit sends the filtered image data stored in the SRAM built in the low-pass filter to the SRAM located in the wireless endoscope capsule to be filtered for all image data Finally, the filtered image data is sent to a JPEG-LS encoder built in the wireless endoscope capsule; Step 16: The JPEG-LS encoder performs lossless compression on all received filtered image data according to JPEG-LS lossless compression encoding under the control of the JPEG-LS compression encoding control unit. Step 17: After the compression operation described in step 16 is completed, the JPEG-LS encoder stores the compressed image data in the SRAM and sends it to the channel encoding unit, and outputs it to the channel located in the wireless channel after encoding. The wireless transceiver in the speculum capsule wirelessly transmits to the outside of the body; The lossless and quasi-lossless image decompression method of the wireless endoscope system is carried out according to the following steps under the control of the JPEG-LS lossless decoding control unit outside the body: Step 21: After the in vitro wireless transceiver device receives the filtered and compressed data from the corresponding device in the wireless endoscope capsule, the data is sent to a JPEG-LS decoder for decoding; Step 22: After the JPEG-LS decoder decodes the compressed data in step 21, send the G component and the B and R components to the G component reconstruction filter and the B and R component reconstruction filter respectively for reconstruction filtering ; Step 23: The reconstruction filter described in step 22 performs reconstruction filtering on the G component and the B and R components respectively according to the following steps: Step 1: the reconstruction filter sends data that does not need reconstruction filtering in the set ROI region to the built-in cache SRAM in the reconstruction filter; for data that needs reconstruction filtering, according to the set q The value is followed by the following vertical reconstruction filtering and horizontal reconstruction filtering; Step II: vertical reconstruction filtering: carry out sequentially from bottom to top according to the following formula, and restore the vertical original data; Image data value of line m after reconstruction = 2* original image data value of line m-original image data value of line m-1, Image data value of line m-1 after reconstruction=2*original image data value of line m-1-original image data value of line m-2, ... Image data value of the second row after reconstruction=2*original image data value of the second row-original image data value of the first row, Image data value of the first line after reconstruction = original image data value of the first line, The third step: horizontal reconstruction filtering: the new data obtained in the second step is carried out from right to left in sequence according to the following formula, to complete the recovery of the horizontal direction data; Image data value of column m after reconstruction = 2* original image data value of column m-original image data value of column m-1, Image data value of the m-1th column after reconstruction=2*original image data value of the m-1th column-original image data value of the m-2th column, ... Image data value of the second column after reconstruction = 2* original image data value of the second column - original image data value of the first column, Image data value of column 1 after reconstruction = original image data value of column 1; Step IV: send the image data after reconstruction filtering to the built-in SRAM; Step 24: the reconstruction filter sends the G component data after reconstruction filtering to a rhombus-to-rectangular converter for transformation; Step 25: Send the reconstructed and filtered B and R component data obtained in step 23 and the transformed G component data obtained in step 24 to an adder for addition to obtain the restored original Bayer color image data; Step 26: send the data obtained in step 25 to a color interpolation processor to interpolate according to the Laplace method to obtain full-color image data. 2. The wireless endoscope system lossless and quasi-lossless image compression method according to claim 1, characterized in that: the lossless compression coding can be JPEG-LS, JPEG2000 lossless compression part and FELICS algorithm any type.

Images