|
With the advances in communication technology and the growing range of communication services, digital video information easy to understand, the transmission quality is good, strong anti-interference ability, high reliability and easy encryption etc. widely used, but the digital information data volume, occupancy band too wide, the storage space is large, so the compression of the digital video image is imperative. The orthogonal transform coding is based on the most image in the DC and low-frequency area is accounted for most of the high-frequency region and a small part of the characteristics, the the airspace image signal transformed into the frequency domain, and an effective method of compressing. Because the eigenvectors of the covariance matrix of the base vectors and a natural image is quite approximate discrete cosine transform is a minimum mean square error criterion quasi best transform coding method, and the DCT is a real number operation, a fast algorithm, which is widely used in compressed video international standards. DCT transform using the correlation between pixels of the image block to eliminate redundancy, theoretically speaking, the response to the entire image a unified transform In order to fully utilize the potential correlation between the presence of all image pixels, is very large, but as the amount of calculation. So the image is usually divided into many small transform the shape of the block is generally square, for ease of fast algorithm, generally the block length selected as the n-th power of 2. Generally considered within the range of 8 × 8-pixel correlation is relatively large, it is generally fixed block size is selected to be 8. But have different statistical properties of the different regions of the image, if the image content within a region very similar, then the correlation of the pixels in this region is very high, the use of the large size of the transform has a better energy concentration; Conversely, if a The more complex the image content within the region, then the correlation between the pixels is relatively weak, the use of large-size transform quantization will produce ringing effect, the use of small size can transform quantization retain more detail, which requires adaptive technique supported. When the average number of bits per pixel is less than 1, the use of adaptive coding technology will play a considerable role. Adaptive Transform Nuclear is the image based on the complexity of the content is divided into a rectangular block, and a corresponding block size of the orthogonal transform and adaptive. DCT-block partition size selection for 4,8 and 16 three blocks is more appropriate, another multi-size encoding mode, the available size is best not more than 3 kinds. 3D-DCT is considered to be the motion compensation alternative technologies, 3D-DCT can use multi interframe correlation to eliminate redundancy therebetween, and the motion compensation techniques can only eliminate the redundancy of up to two frames; addition, as the 3D-DCT- structure is non-recursive, which avoids infinite propagation of transmission errors. In certain image sequence compression performance, comparable transform coding and motion compensation technology. 3D-DCT is very effective when the total amount of movement is low. 3D-DCT longer coding delay and need a lot of storage space, beginning the study does not, but with the rapid development of technology of computer hardware, computer operation speed faster and faster for 3D-DCT Research is also more and more. But at this stage of the 3D-DCT presence of the following problems: 1, no unified guidelines multidimensional matrix calculation method, 3D-DCT matrix operations are very scarce. Adaptive block segmentation criterion confusion. 3, three-dimensional quantitative specific visual model reference. 4, for the transform coefficients in the three-dimensional scanning, RL-VLC entropy coding is basically used in the conventional JPEG method. Problem, Sang Aijun proposed multi-dimensional vector matrix theory is a theory of multidimensional matrix operations, follow the example of the two-dimensional form of matrix-matrix operations extended to multi-dimensional space has a concise representation, the form of easy to understand, The computational complexity is affordable, easy to promote the advantages of the high-dimensional space. Orthogonal transformation of multi-dimensional matrix Hutie root and the multi-dimensional vector matrix theory is applied to color images compression field, and derive the 3D-DCT multi-dimensional vector matrix transform nuclear matrix. The problem, in order to reduce the computational complexity, the paper proposes fast method of segmentation based image activity measure. Gradient-based image activity measure is the best image activity measure, and not only to distinguish between different aspects of the image is very effective and significant correlation and PSNR DCT is horizontal, the vertical direction of the energy is concentrated Therefore, the horizontal and vertical gradient can be a good reaction DCT ability to concentrate energy in that direction, that is, the compression capabilities. Third, the key factors to be considered by the quantizer design problem are: 1, after conversion coefficient distribution of the probability distribution function, and contribute to the minimal distortion of the design of the quantizer. 2, according to the modulation transfer function of the human visual system model of a first order low-pass. The above combination of both to quantization matrix structure. The problem, JPEG RL-VLC revised proposed a new side of a LL-VLC entropy coding method, and a simple context model construction code table is moderate, and there is good potential to promote applied higher dimensional video image entropy coding system. This paper illustrates the effectiveness of multi-dimensional vector matrix orthogonal transform coding system in the field of image compression to some extent, but the system is shorter, there are many imperfections and areas for improvement, great space for development. For example, the multi-dimensional vector matrix discrete cosine transform fast algorithm, as well as with existing popular compression international standards further combined with more integration of new technologies, the need to further study and research.
|