| With the rapid development of digital communications, Internet technology and information industry, multimedia communication in people's daily life plays an important role. Image information storage, transmission becomes more and more frequently, at the same time, the range of applications is growing. Image with intuitive, concrete, vivid features and contains a wealth of information by people as the primary means of access to information, and color images with rich color structure, a more realistic reflection of the real world, so how to compress the mass of color image information effectively is one of the new main research today.For color still images, currently the more common method of study is change the R, G, B signals into Y, U, V signal, and use the same or similar coding methods of image processing for each component independently. Actual studies have shown that R, G, B color components each have a strong correlation between, after the Y, U, V conversion, although the correlation has been weakened between the various components of luminance and chrominance,there are still a lot of visual redundancy. Encoding the color image R, G, B components independently does not take a variety of redundant information in color images into consideration. Therefore did not play the potential of color image compression inner and the compression efficiency is not satisfactory.As we known, in a color image pixel lattice, three primary colors (RGB) for the same physical model and between the various components not only have the same texture, edge and gray gradient but also almost reflect the all of the information outside the color. Between them there is a strong visual similarity, that is, redundant of color space. The traditional coding method is to compress each component separately, can not be removed the redundant between the various components of the association. Therefore, we believe that the key for color image compression coding method is using the same model to express all the information of color images. Fully in to account the spatial structure of color images as well as the various components which exist large number of redundant information. So that the compression ratio and peak signal to noise ratio and other performance are improved.For these problems which exist in conventional color image compression encoding process, this paper reference the results of laboratory and using three-dimensional matrix model to represent color images. R, G, B component of color images are constructed in the same model. Make the image relationship and the spatial structure of the relationship between the various components of unified lower and weight of each component within the correlation between each parts. Color images in order to take full advantage of these features to more effectively remove all kinds of redundancy.Next, the color image is segmented by three-dimensional matrix and be divided into 8 * 8 * 3 sub-matrix. Then, change the sub-array by 4-D vector DCT orthogonal transform and get the transformed coefficient matrix. According to the distribution of data, using Linear non-uniform scalar quantization method to quantify the coefficient matrix. Scan data with many zero coefficients, so a run-length encoding data after the scan code is needed. Finally, using Huffman coding and arithmetic coding transform the data obtained by run-length coding. Experimental results show that the algorithm encoded by this experiment, the image compression ratio and peak signal to noise ratio have been improved to some extent, also verified the validity of this experiment.Finally, in the Windows operating system software using Visual C + +6.0 on programming this algorithm, the experimental results with the current standard JPEG basic methods were compared, the results of the basic method is better than JPEG. Explains to some extent, the effectiveness of the algorithm, but also illustrates the matrix theory of multi-dimensional vector processing in image compression has some advantages and potential can be. |
