Image compression coding technology

Predictive coding

Predictive coding is based on the strong spatial and temporal correlation between image and video data. The current pixel value is predicted by the neighbor pixel values with strong correlation with the current pixel, and then the difference between the current pixel value and the predicted value is quantized and encoded.

The stronger the correlation between the neighbor pixel and the current pixel is, the closer the predicted value is to the current pixel value, the smaller the prediction error is, the less coding bits are needed under the same conditions, and the higher the compression efficiency is.

Therefore, the predictive coding core is the design of the predictor. In practical applications, the best probability prediction algorithm is usually determined according to the statistical probability distribution of the image, or the adaptive predictor is used according to the local characteristics of the image data to enhance the prediction effect. Predictive coding is divided into linear and nonlinear prediction, or intra and inter prediction.

Entropy coding

Entropy coding is based on the probability distribution characteristics of image data. It transforms the sequence symbols representing image pixels into a compressed bit stream for transmission and storage. The basic idea is that the pixel values with large occurrence probability are represented by short codes, and the pixel values with small occurrence probability are represented by long codes. It is proved that the average length of the output codeword is the shortest, which is close to the entropy of the source.

Statistical coding is lossless data compression coding, commonly used Huffman coding, arithmetic coding and run-length coding. These entropy coding algorithms in different combinations, such as Huffman coding combined with run-length coding, arithmetic coding combined with run-length coding, etc., are used as further coding after coding algorithms such as transform and prediction.

Transform coding

Transform coding was proposed by H. Andrews et al in 1968. It transforms an image from one representation space to another representation space by performing some function transformation on the image, and then eliminates the high frequency components with small energy and less sensitive to human eyes by quantization. The basic process of transform coding is that the image is firstly transformed by some form of orthogonal transformation to obtain the transform coefficient matrix, and then the coefficient matrix is quantized with different precision according to the human visual characteristics, and finally the quantized coefficient matrix is encoded by entropy coding.

Your task

Choose a coding technique that interests you and talk about it.

Share your thoughts and ideas in the comments below.