RU2488885C1 - Method to process raster images - Google Patents

Abstract

FIELD: information technologies.

SUBSTANCE: in the method of processing of raster images, including compression of an image by the method of "cut block coding" or its modifications, before the procedure of compressing coding they perform digital filtration, which increases sharpness of the compressed image, and after the decoding procedure they perform smoothing digital filtration of the decoded image.

EFFECT: improved quality of decoded raster images when methods used for their compressing coding on the basis of cut block coding or their modifications, improved current formalised criteria.

3 cl, 8 dwg

Description

The invention relates to the field of processing bitmap images, and in particular to methods of compression encoding and decoding for communication systems and storing bitmap images.

The invention can be used in data transmission and storage systems, in computer technology, in instrumentation, in telemetry systems.

A known method of processing raster images in which the encoding of the image for compression is based on the JPG method (See application for US patent US 2007/0036451 A1).

The disadvantage of this method is that it does not provide a guaranteed, constant and predetermined compression ratio for the entire compressible image as a whole, and for each image fragment of a given size. Another disadvantage is that such data does not have a high resistance to loss and distortion during transmission. Distortion of compressed data corresponding to a certain area of the image leads to distortion of other areas. Both when compressing images using the JPEG method, and during recovery, the required number of elementary operations per pixel is quite large. In this case, it is impossible to effectively parallelize the compression procedure, since all blocks are not independent.

There is also a known method of processing bitmap images, in which the image is encoded for compression using the truncated block coding method (PTS - “block truncation coding” in the English language literature). (See application for US patent US 2010/0067812 A1). It assumes that the image is divided into blocks of a given size. Moreover, the blocks can be of arbitrary size and shape, not necessarily square, and not necessarily they must have the same size for the entire image.

This method is taken as a prototype of the proposed invention.

The disadvantage of this method is that the quality of the image recovered after compression according to commonly used formalized criteria, such as peak signal-to-noise ratio (SNR or PSNR in the English literature), standard deviation (MSE or MSE in the English literature), structural similarity (SSim in the English literature ) and others, when using blocks of larger size than 4 × 4 pixels, it turns out to be significantly worse than when compressing the same images using JPEG with the same compression ratio. Reducing the block size improves the quality of the recovered images, but at the same time sharply reduces the compression ratio.

In the proposed method, the technical task is to improve the quality of decoded raster images when using methods for compressing them based on truncated block coding or their modifications.

This technical problem is solved due to the fact that in the method of processing raster images, including image compression by the method of "truncated block coding" or its modifications, digital filtering is performed before the compression coding procedure, which increases the sharpness of the compressed image, and after the decoding procedure, smoothing digital filtering of the decoded Images. Filtering can be applied either to all pixels of the compressed image, or selectively, for example, sharpening filtering for low-contrast blocks, smoothing filtering for blocks with high contrast. Filtering can also be adaptive, explicitly or indirectly taking into account the distribution of pixels by brightness in the vicinity of the filtered pixel.

This method is illustrated with the help of the attached drawings on the example of selecting the parameters of matched filters with a window size of 3 × 3, taking into account that both sharpening (pre-filtering stage) and smoothing filters (post-filtering stage) must be symmetrical, i.e. multipliers for pixels located at a given distance from the current (filtered) should be the same regardless of direction. 1 shows a 3 × 3 block during prefiltration. Figure 2 shows the block 3 × 3 during post-filtering. Figure 3 in the form of a three-dimensional graph shows the results of a model program for selecting filter parameters for a conventional LENA reference image when truncated block coding is used with a block size of 3 × 3 and the number of quantization levels equal to two.

Figure 4 shows the original 3 × 3 block in another example of the use of matched filters with respect to the 3 × 3 pixel block, when the prefiltration and post-filtering procedures also use environmental pixels outside the borders of the block. Figure 5 shows the decoded block after applying the truncated block coding method without filtering. Figure 6 shows the source block after prefiltration. 7 shows a decoded block after a truncated block coding method with prefiltration. Fig. 8 shows a decoded block after post-filtering.

и среднеквадратичное отклонение яркости σ, после чего формируется битовая матрица, в которой каждому пикселу блока, яркость которого превосходит порог $$\overline{u}$$

, ставится в соответствие один бит (для определенности - 0, если яркость пиксела меньше $$\overline{u}$$

, 1 - в противном случае). Для каждого блока кодер сохраняет $$\overline{u}$$

, σ и построенную описанным выше методом битовую матрицу, фактически содержащую квантованные на два уровня значения яркостей пикселов блока. При восстановлении пикселы, которым в битовой матрице соответствовали значения 1 и 0, заменяются наIn the classical implementation of the truncated block coding method, the average brightness is calculated for each image block $$\overline{u}$$

and standard deviation of brightness σ, after which a bit matrix is formed in which each pixel of the block, the brightness of which exceeds a threshold $$\overline{u}$$

, one bit is mapped (for definiteness, 0 if the pixel brightness is less $$\overline{u}$$

, 1 - otherwise). For each block, the encoder saves $$\overline{u}$$

, σ and the bit matrix constructed by the method described above, which actually contains the brightness values of the pixel pixels of the block quantized by two levels. When restoring, pixels to which the values of 1 and 0 correspond in the bit matrix are replaced by

и $$\overline{u}-\sigma \sqrt{\frac{q}{m-q}}$$

$$\overline{u}+\sigma \sqrt{\frac{m-q}{q}}$$

and $$\overline{u}-\sigma \sqrt{\frac{q}{m-q}}$$

(here m is the number of pixels in the block, and q is the number of pixels whose brightness exceeds the threshold value, i.e., those to which 1 corresponds in the bit matrix). This approach allows for each block of the reconstructed image to maintain unchanged the average brightness and dispersion.

The images restored after compression by the truncated block coding method contain a number of visual artifacts associated with a significant (up to 2, 4 or 8, depending on the number of quantization levels) reduction in the number of gradations of image brightness within each coded block and, as a result, with smoothing of the intra-block relief.

The compression coefficient when using the truncated block coding method is determined only by the block size, does not depend on the image features, and at 8 quantization levels it is usually from 2 (4 × 4 pixel block) to 2.5 (8 × 8 block).

In the proposed method for processing raster images before the compression encoding procedure, digital filtering is performed to increase the sharpness of the compressed image, and after the decoding procedure, smoothing digital filtering of the decoded image is performed.

Here are examples of the method.

Example 1

(A pair of matched filters is applied to the entire image, to each pixel, regardless of its brightness and the brightness of the surrounding pixels).

Let us consider an example of selecting parameters for matched filters with a window size of 3 × 3, taking into account that both sharpening and smoothing filters are symmetrical, i.e. multipliers for pixels located at a given distance from the current (filtered) should be the same regardless of direction.

In the first example: m0 (w0) is the factor applied to the current filtered pixel, m ₁ and m ₂ (w ₁ and w ₂ ) are the factors applied to pixels one vertical position from the current one, in horizontal, and diagonal directions. Since the average image brightness when using filters should not change, for sharpening and smoothing filters, the conditions are met, respectively

m ₀ -4 * (m ₁ + m ₂ ) = 1

and

w ₀ + 4 * (w ₁ + w ₂ ) = 1.

Thus, each filter is completely described by just a pair of coefficients, and the experimental estimation of the filter parameters is reduced to the selection of pairs (m ₁ , m ₂ ) and (w ₁ , w ₂ ) that provide a minimum standard deviation (MSE), defined as

$$MSE=\frac{\mathrm{one}}{N\times M}{{\displaystyle \sum _{j=\mathrm{one}}^N{\displaystyle \sum _{i=\mathrm{one}}^M\left(I_{i,j}-{\stackrel{⌢}{I}}_{i,j}\right)}}}^2$$

- соответственно пикселы исходного и восстановленного изображений с координатами i и j, а N и М - длина пиксельной строки и число строк изображения.where I _{i, j} and $${\stackrel{⌢}{I}}_{i,j}$$

- respectively, the pixels of the original and restored images with coordinates i and j, and N and M are the length of the pixel line and the number of lines of the image.

When using a simulation program to search for pairs of matched filters for each sharpening filter used for pre-filtering, an appropriate (i.e., ensuring the minimum standard deviation for a given sharpening filter value) post-filter is selected.

Figure 3 in the form of a three-dimensional graph shows the results of the work of the model program for selecting filter parameters for the LENA image when encoded by the method of truncated block coding with a block size of 3 × 3 and the number of quantization levels equal to two. The horizontal plane corresponds to different values of the pairs (m ₁ , m ₂ ), the lowest values of the standard deviation corresponding to each pair are plotted vertically. The graph is a concave surface with a minimum at a point with coordinates m ₁ = 17/64, m ₂ = 3/64.

The use of a sharpening filter with the indicated parameters together with a smoothing filter matched with it with w ₁ = 26/256, w ₂ = 11/256 allows for compression using a truncated block coding method with a block size of 3 × 3 and the number of quantization levels 2 for obtaining a test image LENA MSE = 8.464. Without filtering, the standard deviation for the same image encoded using a block of the same size and the same number of quantization levels is 18.196.

Thus, the use of matched filtering ensured a gain in MSE of 2.15 times, which corresponds to an increase in the peak signal-to-noise ratio by 3.324 dB, without increasing the amount of data transmitted to the decoder.

Example 2

In the second example of using matched filters with respect to the 3 × 3 pixel block and the prefiltering and postfiltering procedures, the environment pixels outside the block are also used.

When using the truncated block coding method without filtering, the standard deviation for a given decoded block (not the whole image, as in Example 1) is 65.78, and when using filtering, it is 18.89.

The claimed method of processing raster images can be carried out in industry using advanced technologies, materials and computational processes and can be used in data transmission and storage systems.

The proposed method for processing bitmap images was tested experimentally at Vokord SoftLab LLC. Experimental verification of this method showed an increase in the quality of decoded raster images when using methods for their compression coding based on truncated block coding or their modifications.

Experiments carried out on a series of test images using sharpening and smoothing filters with a constant window of 3 × 3 pixels have demonstrated that the application of this approach allows reducing the standard deviation by an average of 2-4 times. The parameters of filters of this type are determined by the block size and the number of quantization levels and depend little on the nature of the compressed image. This allows you to use the same pair of filters to compress a wide variety of images, which is confirmed by experiments. The application of the described approach is possible using any modification of the truncated block coding method.

It turned out that the implementation features of the filters used for preliminary filtering of encoded images and post-filtering of the restored images are not critical. Essential for the filter applied to the pixels of the original image is its ability to emphasize the relief elements of the image block, i.e. increase the difference between the minimum and maximum values of the brightness of the pixels of the encoded block, and for the filter used in the post-filtering stage, the ability to smooth these elements, i.e. reduce the difference between the minimum and maximum values of the brightness of the pixels of the decoded block. Filters may not be applied to all image pixels, but selectively, to those meeting established criteria, for example, to those in low-contrast blocks. Filters can also be adaptive, explicitly or indirectly taking into account the distribution of pixels by brightness in the vicinity of the filtered pixel.

The proposed method for processing raster images is recommended for use in data transmission and storage systems, in computer technology, in instrument making, and in telemetry systems.

Claims (3)

1. A method for processing raster images, including image compression by the method of "truncated block coding" or its modification, characterized in that before the compression encoding procedure, digital filtering is performed to increase the sharpness of the compressed image, and after the decoding procedure, smoothing digital filtering of the decoded image is performed. 2. The method of processing bitmap images according to claim 1, characterized in that sharpening filtering and smoothing filtering may not be applied to all image pixels. 3. The method of processing raster images according to claim 1, characterized in that the sharpening filtering and smoothing filtering can be adaptive, explicitly or indirectly taking into account the distribution of pixels by brightness in the vicinity of the filtered pixel.

Images