CN1448892A - Compression rate pre-allocation algorithm for JPEG 2000 multi-picture photo - Google Patents

Abstract

The invention discloses a compression rate pre-allocation algorithm among JPEG2000 image slices. It can effectively distribute the compression rate set by the user to each image slice according to the amount of information contained in each image slice in JPEG2000 processing, thus solving the problem of simply allocating the overall compression rate to each image slice Poor chroma and rough edges. And the JPEG-LS and PNG prediction templates and edge detection operators are introduced to calculate the information content of image slices. A large number of experimental results show that: using the compression rate pre-allocation algorithm, the visual effect of the reconstructed image can be improved and strengthened to varying degrees, and the MSE is significantly reduced. It is especially effective for images with concentrated regions; JPEG-LS and PNG are used to predict The algorithm and the edge detection operator of the small template to measure the information of the image slice have better real-time performance and effectiveness, and can meet the hardware requirements.

Description

A Pre-allocation Algorithm of Compression Ratio for JPEG2000 Multi-image Slices

1. Technical field

The invention belongs to the technical field of VLSI design. Specifically, it involves designing a new JPEG2000 compression rate pre-allocation algorithm in JPEG2000 hardware implementation.

2. Background technology

JPEG2000 static image compression standard has been a hot spot in engineering and academic circles since its release. All the compression rate control (rate control) parts in the JPEG2000 protocol are carried out for an image slice, which explains in detail how to allocate a finite length of codewords in an image slice in order to minimize the distortion of the image after reconstruction. However, for a relatively large image, it is generally divided into multiple image slices for processing, and all forward component transformation, DC displacement, discrete wavelet transform, quantization, arithmetic coding, and layered truncation and packaging will only be performed in one Image slices are made internally. So how should the compression ratio be allocated between image slices? For this problem, from all the JPEG2000 protocols that have been released domestically and internationally, and the papers on JPEG2000 compression rate control, as well as the realization of JPEG2000 hardware and software structure, there is no relevant literature discussing this problem, let alone put forward related solutions. Judging from the JasPer/JPEG2000 source program, it does not take into account the information difference between the image slices, but just distributes the overall compression rate of the image to each image slice on average. After observation and research, the applicant found that this processing method is not satisfactory when the information is evenly distributed on the entire image or the compression ratio is relatively low; but if the information is extremely unevenly distributed among various image slices and the compression ratio is Under normal circumstances, when using this method to compress images, there will be obvious chromaticity difference and rough edges between image slices. Of course, the user cannot manually set the compression rate of each image slice. There should be an adaptive algorithm that automatically allocates the compression rate according to the information content of the image slice to solve this problem.

However, the traditional evaluation methods of information content, such as Shannon information content, need to count the pixel values of the image, and also perform logarithmic operations. The same MSE statistical method requires two scans of the original image and a square operation. Therefore, the traditional calculation method of information content cannot meet the real-time requirements of hardware due to its large amount of calculation. Therefore, it is necessary to find a suitable statistical method of information content to evaluate the information content of images.

3. Contents of the invention

According to the defects and deficiencies in the above-mentioned background technology, the object of the present invention is to provide a JPEG2000 multi-picture slice compression rate pre-allocation algorithm.

The solution adopted by the present invention is: firstly calculate the information content of the image, and then pre-allocate the compression rate between the image slices according to the information content of the image; in addition, introduce JPEG-LS and PNG prediction module and edge detection module to measure the image information content;

Include at least the following steps:

1) Introduce prediction templates and edge detection operators to evaluate the information content of image slices

a) Adopt JPEG-LS or PNG prediction module to predict within the scope of the four neighborhoods, let x' be the predicted amount, then predict the information content: x-x' (x is the actual pixel value), the information content of the overall image is the sum of the predicted information content of all pixels;

B) the method for edge detection operator evaluation image information volume: at first act on original image with edge detection template, obtain the information volume of each pixel; The sum of each pixel information volume is exactly the information volume of image;

2) Assign the compression ratio of each image slice according to the product of the information content of different image slices and the compression ratio set by the user as a weighting coefficient.

The JPEG2000 multi-image slice compression rate pre-allocation algorithm of the present invention can improve and strengthen the visual effect of the reconstructed image in different degrees, and the MSE is significantly reduced. It is especially effective for images with regional concentration of the target; use JPEG-LS and PNG to predict The algorithm and the edge detection operator of the small template to measure the information of the image slice have better real-time performance and effectiveness, and can meet the requirements of hardware design.

4. Description of drawings

Fig. 1 is a schematic diagram of the prediction process of the present invention;

Fig. 2 is a schematic diagram of a prediction template of the present invention;

Fig. 3 is a schematic diagram of the JPEG-LS prediction algorithm;

Fig. 4 is a schematic diagram of PNG prediction algorithm;

Fig. 5 is a schematic diagram of the edge detection process;

Fig. 6 is a schematic diagram of a simplified edge detection template;

Fig. 7 is an improved JPEG2000 encoding flow chart.

5. Specific implementation

The invention is an algorithm for automatically distributing the compression rate among the image slices according to the information content of the image slices. The present invention will be described in further detail below in conjunction with the accompanying drawings.

1. Evaluation method of information content

There are many evaluation methods for image information content, and Shannon information content is usually used to measure it. However, the amount of Shannon information needs to count the pixel values of all points, and also perform logarithmic operation on the obtained statistical probability, which requires a large amount of calculation and poor real-time performance. The same mean square error (MSE) MSE is also a method to measure the information content of the image. It needs to scan the entire image twice, and also perform square operations, so the calculation amount is quite large, and more importantly, it cannot be realized. Parallel operation. In addition, Shannon's information content and MSE statistical methods do not consider the domain information of image pixels. Aiming at the shortcomings of the above two methods, it is the first time to introduce edge detection and prediction templates to measure the practical algorithm of image information content. The real-time performance and effectiveness of these two operators are much better than those of Shannon and MSE.

*Prediction information content Add a prediction module to the image, then use the difference between the predicted value and the source pixel value, and finally use this difference to calculate the information content.

The calculation process of predicting information content is shown in Figure 1. This prediction module is mainly based on four domains (Fig. 2). The usual prediction modules include JPEG-LS (Fig. 3) and PNG (Fig. 4). A prediction module has three favorable characteristics:

1) The amount of calculation is small;

2) Domain information is considered;

3) It is more in line with the definition of information volume.

*Statistics based on edge detection

The edge is a very important feature for an image, so the edge is also an important symbol to measure the information content of the image. It is also an important method to introduce some edge detection templates to first detect the edge of the image, and then use the edge to determine the information content of the image.

Common edge detection operators include Roberts intersection operator, Sobel operator, Prewitt operator, Laplace operator, LOG operator and Canny operator, etc. The richer the edge, the greater the information content contained in the image. The best operator to check the edge is the Canny operator, which can not only detect the edge of the image, but also has the function of denoising. It is found through experiments that after using the 9×9Canny operator for information content statistics, the average effect of the compression rate distribution is indeed the best, and the stability is high. The fly in the ointment is that the amount of calculation is too large, whether it is hardware or software, I am afraid that the overhead will be unbearable.

The present invention recommends a simplified edge detection operator (as shown in FIG. 6 ). The use of this template does not require multiplication, and the efficiency is greatly improved. In practice, it is also possible to perform interval sampling every 4 points or 8 points, and then perform edge detection. This not only improves the computing speed, but also can detect large edges and is not easily disturbed by noise. Using this method only needs to pay a small calculation cost, and the image quality can be significantly improved under the condition of constant compression ratio, which is a compression ratio pre-allocation algorithm recommended by the present invention.

2. Pre-allocation process

Let ImageSize be the size of the original image, N be the number of divided image slices, ratio be the compression ratio given by the user, and ri (i=1, 2, 3, ... N) be the compression rate weight coefficient of each image slice, Satisfy the formula (2.1): $\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\mathrm{the\; ri}=1---\left(2.1\right)$

Let R _i be the number of bytes allocated to each image slice, then:

(2.2)

Ri=ImageSize×ratio×ri Here, the compression rate weight coefficient r _k can be obtained by formula (2.3). $r_k=\frac{I_k}{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{I}_i}---\left(2.3\right)$

Among them, I _k is the information content corresponding to image slice k, which can be calculated by any evaluation method given above. The JPEG2000 encoding process after the compression rate pre-allocation module is added is shown in Figure 7 .

Using the pre-allocation algorithm based on the information content of image slices to test many images with regional concentration, the results show that the effect of compression and decompression can be greatly enhanced, and this algorithm has high real-time and effectiveness.

The information content of an image slice refers to the minimum compressed storage space required to reconstruct an image slice. There is no exact quantitative standard for the latter meaning, and it is closely related to specific algorithms for removing redundant information. Therefore, for an image, it is not only related to the pixel value of the image, but also closely related to the position arrangement of the pixels. However, Shannon's information content and MSE statistics do not take into account the location information of pixels. Both the prediction template and the edge detection module take into account the relationship with domain information, so this information content not only expresses the statistics of pixel values, but more importantly, it includes the accumulation of pixel position information. In addition, the calculation amount of the prediction template and the edge detection operator is small, and the information amount of the image slice can be calculated in real time by adding 2 to 3 image line stores at the front end of the image input without scanning the image multiple times.

The compression rate pre-allocation algorithm refers to a method of allocating the compression rate set by the user to each image slice according to the size of the information content of each image slice. First, calculate the information content of each image slice according to the method of information evaluation, and then assign different compression ratio distribution weight coefficients to each image slice according to the difference in information content, and then multiply it with the compression rate set by the user. Get the actual compression rate of the current image slice.

A large number of experimental results show that: using the compression rate pre-allocation algorithm, the visual effect of the reconstructed image can be improved and strengthened to varying degrees, and the MSE is significantly reduced. It is especially effective for images with concentrated regions; use JPEG-LS and PNG to predict The algorithm and the edge detection operator of the small template to measure the information of the image slice have better real-time performance and effectiveness, and can meet the requirements of hardware design.

Claims (2)

1. A kind of JPEG2000 multi-picture sheet compression rate pre-allocation algorithm, it is characterized in that, at first calculate the information content of image, then carry out the pre-allocation of compression rate between image slices according to the information content of image; Introduce JPEG-LS and PNG in addition The prediction module and edge detection module to measure the information content of the image; Include at least the following steps: 1) Introduce prediction templates and edge detection operators to evaluate the information content of image slices; a) Adopt JPEG-LS or PNG prediction module to predict within the scope of the four neighborhoods, let x' be the predicted amount, then predict the information content: x-x' (x is the actual pixel value), the information content of the overall image is the sum of the predicted information content of all pixels; B) the method for edge detection operator to evaluate the information volume of image is: at first act on original image with edge detection template, obtain the information volume of each pixel; The sum of each pixel information volume is exactly the information volume of image; 2) Assign the compression ratio of each image slice according to the product of the information content of different image slices and the compression ratio set by the user as a weighting coefficient. 2. JPEG2000 as claimed in claim 1 multi-picture compression rate pre-allocation algorithm, is characterized in that, described pre-allocation process is carried out by following formula: Let ImageSize be the size of the original image, N be the number of divided image slices, ratio be the compression ratio given by the user, and ri (i=1, 2, 3, ... N) be the compression rate weight coefficient of each image slice, Satisfy the formula (2.1): $\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}\mathrm{the\; ri}=1---\left(2.1\right)$ Let R _i be the number of bytes allocated to each image slice, then: (2.2)        Ri＝ImageSize×ratio×ri Here the compression rate weight coefficient r _k can be obtained by formula (2.3); $r_k=\frac{I_k}{\underset{i=0}{\overset{N-1}{\mathrm{\Σ}}}{I}_i}---\left(2.3\right)$ Among them, I _k is the information content corresponding to image slice k, which can be calculated by any evaluation method given above.