Image super-resolution method based on SAE and sparse representation

Abstract

The invention discloses an image super-resolution method based on SAE and sparse representation, and belongs to the field of image processing. The image super-resolution method mainly comprises an off-line training stage and a test refactoring stage, wherein in the off-line training stage, image characteristics extracted by an SAE (Sparse Auto Encoding) model are subjected to dictionary training, and a dictionary pair reflecting corresponding relations of high-resolution images and low-resolution images is established; in the test refactoring stage, low-resolution images inputted by a user are subjected to super-resolution reconstruction by the obtained dictionaries and a sparse representation method. Through the application of the image super-resolution method, unsupervised learning training is performed on original image sampling data by using the SAE model, so that the defects that manually designed operator characteristic extraction is time-consuming and strenuous and the extracted characteristics are single are avoided, meanwhile, image characteristics represented by SAE compression are directly used for training of the high-low-resolution dictionary pair, the dictionary training is facilitated, lost detail components in the images can be estimated by the sparse representation method, and higher-quality high-resolution images can be restored from the low-resolution images conveniently.

Description

technical field [0001] The invention belongs to the field of image processing and relates to an image super-resolution method based on SAE and sparse representation. Background technique [0002] With the development of science and technology and the continuous strengthening of social security awareness, video surveillance systems are widely used in the field of public security. High-quality, high-resolution images are crucial to the realization of video surveillance system functions because they can reflect more detailed information in visual scenes and people. However, in practical applications, the generation of low-resolution images is unavoidable. On the one hand, many imaging systems (such as infrared imagers, CCDs, CMOS sensors, etc.) are limited by their inherent sensor array arrangement density and affected by factors such as target movement, poor focus, and system noise. It is difficult to achieve the desired level, with defects such as blurring, noise, and defor...

AI Technical Summary

Problems solved by technology

[0007] However, traditional sparse representation-based image super-resolution techniques mostly use artificially designed feature extraction operators (such as gradient or Laplacian operators) to extract features from images (blocks). The advantage of this method is that The extracted features can describe the structure and edge features of the image. The disadvantage is that on the one hand, the features extracted by the artificially designed feature extraction operator are fixed and single. On the other hand, this method is easy to increase the dimension of the feature space. During dictionary training, dictionary training is difficult and data overfitting tends to occur. In order to avoid this phenomenon, some data dimensionality reduction methods (such as PCA method, popular dimensionality reduction methods, etc.) are often used to reduce the dimension of the feature space. The features after dimensionality reduction are used for dictionary training. Since manual feature extraction and dimensionality reduction methods are performed independently, this method will damage the reliability of the original feature space description image to a certain extent.

Images