A high-quality image compression method

Abstract

The application provides a high-quality image compression method, which combines a clustering compression algorithm and a deep neural network compression algorithm; an original image is input into the clustering compression algorithm first, the number of clustering clusters K is determined, K-class clustering is performed on pixel values of the image, and output data X is obtained after clustering; at this time, the image compression ratio is n:1; the data X is input into the deep neural network compression algorithm, two-layer encoding and two-layer decoding are performed, the network parameters of each layer are adjusted through a mean square error function to make the error minimum or change very small; the final output data of the encoding layer is H1; the final output data of the decoding layer is X 1 , X 1 is also the output of the entire image compression algorithm; X 1 and X are input into the mean square error function to obtain the error L1 of the entire image compression method; the compression ratio of the deep neural network part is m:1, and the compression quality of the compressed picture is improved through this step of compression; the compression ratio of the entire algorithm is nm:1.

Images