Block adaptive carton image compression method based on dictionary learning

Abstract

The invention relates to the technical field of image processing, and discloses a dictionary learning-based block adaptive carton image compression method, which comprises the following steps of: performing coding compression by using a learning dictionary, image segmentation, edge detection, adaptive error setting and a quadratic optimization sparse coefficient matrix; the method comprises the following steps: training an offline dictionary for a carton image sample set by using a K-SVD (K-Singular Value Decomposition) algorithm, carrying out improved Canny edge detection after image partitioning to obtain a contour area, determining the structural complexity of an image block, adaptively setting an error of a sparse representation model in a partitioning manner, calculating an initial sparse coefficient matrix of the image block under an original error by using an OMP (Orthogonal Matching Practice) algorithm, and carrying out sparse representation on the initial sparse coefficient matrix; the method comprises the following steps: firstly, extracting a sparse coefficient matrix according to a self-adaptive error, secondly, carrying out secondary optimization on the sparse coefficient matrix according to the self-adaptive error, finally extracting a non-zero value and an index of the non-zero value, and carrying out coding compression. Image data sparseness and reconstruction can be completed, computer storage resources occupied by image data are reduced, and the storage cost is reduced.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a block adaptive carton image compression method based on dictionary learning. Background technique [0002] With the rapid development of image acquisition technology in monitoring, medicine, remote sensing and other fields, the collected image data has increased significantly, so it is of great significance to study image compression to facilitate image transmission and storage. Since the JPEG image compression standard was proposed It has been widely used in just a few years. At present, the built-in image compression method of most electronic devices used in our daily life is JEPG; [0003] In recent years, compressing images using sparse coefficient matrices and over-complete dictionaries on the basis of compressed sensing has received considerable attention. The existing methods are image compression based on dictionary learning, through improved matching pursuit ...

AI Technical Summary

Problems solved by technology

[0003] In recent years, compressing images using sparse coefficient matrices and over-complete dictionaries on the basis of compressed sensing has received considerable attention. The existing methods are image compression based on dictionary learning, through improved matching pursuit algorithms, optimized learning dictionaries and other methods Improve the sparsity of the sparse matrix to achieve high compression of the image, but the structural characteristics of the image are not fully utilized when solving the sparse coefficient matrix, which is prone to the following deficiencies:

[0004] 1. Taking the carton picture as an example, when reconstructing the carton image, more consideration is given to the text information such as the brand of the goods on the picture, so using the same compression ratio to compare areas with few features, compressed storage is a waste of physical resources;

[0005] 2. Areas with rich image features need to retain more detailed information. As early as 2012, it was proposed to use a targeted dictionary for a specific type of image to adaptively compress the image. The input image is encoded. This scheme is based on a sparse dictionary structure. The compact representation of the structure makes the transmission cost of the dictionary and compressed data relatively low, but this method transmits the dictionary and compressed data together, which increases the image compression rate;

[0006] 3. Subsequently, on the previous basis, someone proposed a multi-view image compression method based on the local patch dictionary, which divides the image into small blocks, applies sparse coding to each small block, and uses dictionary atoms to construct the selected The vector of the region represents overlapping small blocks. This technique does not need to store or transmit dictionary atoms. This method has better compression performance than JPEG2000, but it is only for video images with continuous frames, and cannot adaptively compress images.

Images