Research On Image Fusion Algorithm Based On Compressed Sensing Theory

Image fusion technology is a comprehensive discipline the combining multi-sensor technology and data fusion technology. Through the image fusion technology, the source images of different imaging mechanism or different image properties can be fused, finally gets a picture which the information content is rich and the texture details are more clearer, and it is more help later analysis and processing.The advent of the era of big data, the compressed sensing theory has been widely researched and applied.Similarly, with the rapid development of machine vision, people need to deal with more and more images.The image data dimension reduction has become an important mean to improve processing efficiency.Therefore,the image fusion algorithm based on compressed sensing will become the important research direction.In this paper, according to the excellent characteristics of compressed sensing,the image fusion algorithm based on compressed sensing is studied, the main research contents are as follows:First, researching and analyzing the basic theory of image fusion, which includes the theoretical framework, commonly used transform method as well as the evaluation index, and summarizes the application basis of the indexes. At the same time, analyzing and researching the compressed sensing theory and its three important links: signals sparsity, compressed sensing and reconstruction.Second, according to the characteristics of the non-subsampled contourlet Transform(NSCT), the compressed sensing image fusion algorithm based on NSCT is proposed. The high frequency coefficients after the NSCT transform CS transformation, according to the different characteristics of the high and low frequency coefficients, that respectively use different criteria for fusion. Among them,the coefficients of the high frequency observation can take the pulse coupled neural network method, and the low frequency coefficients uses average fusion rule.Comparing the method with other methods, the result shows that the method can gain good effect whether on the subjective visual or on the objective evaluation index.Third, the fusion algorithm based on image classification is proposed. The method analyzes the application scenarios of image fusion, namely the common fusion method as the basis for the detection and tracking, firstly needs to image fusion and then the image is classified and detected. In order to improve the accuracy of classification detection, this method carries on the classification before the image fusion, it is divided into two parts, the target and the background, which choose different rules for fusion. Among them, in the classification process, in order to improve the classification efficiency, selected the method of the combination of CS and K-Means to classify. Comparing the method with other methods, the result shows the method can get a better effect of image fusion, more helpful to image detection.