Research On Image Set Compression Algorithm

With the development of image capturing technology, the resolution of digital images and the amount of data are increased continuously. So how to compress the digital images effectively is still a very serious problem. Traditional compression algorithms like JPEG and JPEG2000 compress single images mainly using the inter-pixel redundancy, the coding redundancy and the psychovisual redundancy. Since many images are taken in the same or similar scenes, there must be some other redundancies among them. If these redundancies are used effectively, the compression ratio will be further improved and the storage space will be saved. Image set compression algorithms are used for the compression of image set which contains many similar images. Both set redundancy and single image redundancies are used for image set compression.This paper is focused on the image set division problem, complexity problem and efficient compression problem in traditional image set compression algorithms. The main work of this paper includes:(1) This paper introduces the whole development process of two types of image set compression algorithms which are based on representative signals and graph individually. We also introduce some image set compression algorithms briefly;(2) For the image set division problem in lossless image set compression, this paper proposed an image set division algorithm based on the mean of correlation coefficient. The algorithm proposed firstly gets a quadratic curve by the mean of image set correlation coefficients, and then divides the image set based on this quadratic curve before compressing it. This algorithm can ensure the image set compression gain reach the maximum;(3) For the high complexity problem in lossy image set compression, this paper proposed an image set compression algorithm based on the undirected weighted graph. The algorithm proposed firstly down sample the Y-component of all images in the image set and use correlation coefficient as the parameter of edge weight function to construct an undirected weighted graph. Then we calculate the Minimum Spanning Tree (MST) of the graph using Kruskal’s algorithm and rearrange the images based on the depth of the leaf vertex and Breadth First Search (BFS) method. At last, the rearranged images are coded by the latest video coding technique High Efficiency Video Coding (HEVC);(4) Different from the two types of image set compression algorithms, this paper also proposed an image set compression algorithm based on non-negative matrix factorization. The algorithm proposed effectively improves the compression performance through non-negative matrix factorization on image and dictionary matrix shared by different images.