Research On Image Fusion Algorithm Based On Second Generation Curvelet Transform

Multi-scale analysis has been widely used in the field of image processing and developed well. Image fusion can combine information between images which come from different sources to obtain a richer and clearer image. As a multi-scale analysis tool, the second generation Curvelet transform can describe the image curve singularities well. Images can easily be operated on decomposition coefficients by using the Curvelet transform, and get the fused image which has more prominent visual effect. Therefore, The Curvelet transform is introduced into the field of image fusion is feasible and practical.This article is about to research the fusion algorithm based on Curvelet transform, which is designed to provide a better fusion method to ensure the quality of the fused image, and lays the foundation of subsequent image processing. Specific contents and results are as follows:(1) Depending on the image characteristics to improve fusion rules. Considering the focus area has prominent details,we use gradient energy to divide areas, and adopt different rules in different regions based on significant factor. The gap on infrared and visible images target gray distribution in the same scene is a little big, Images can be divided into similar and dissimilar regional area by using this characteristics, then based on similarity and feature to combine information together. Experimental results show that the similarity-based guide better to express the details in visible images and the radiation information in infrared images. Different modal medical images can provide helpful information, and do not overwrite each other. Therefore we consider an image as a main image, and take other modal image information into the main image. The main image is selected according to the information entropy, and the fusion rules is determined by its high frequency coefficient ratio. Experiments show that this method can be useful for integrating information in multi-modal medical images.(2) The edge detection based on Sobel operator is introduced into image fusion and divides the coefficients into the edge and non-edge portion, we can apply different rules to different parts. Processing low-frequency coefficients after Curvelet and wavelet transform decomposition can effectively improve the clarity of the image by using this method Experiments show that the image which is obtained by this method, its details and textures is getting better.(3) Improving the PCNN simplified model parameters. Traditional model sets strength of the link as a fixed value which does not meet the characteristics of the image, therefore, this paper will combine the linking intensity with image characteristics. In multi-focus images, the adaptive linking intensity can be set based on the sharpness; In the infrared and visible images, the adaptive linking intensity can be set according to the regional energy. Then with the local statistical characteristics of the image as an external input, the fusion high frequency coefficients is determined by the number of ignition times. Experimental results show that the improved PCNN simplified model which is combined with Curvelet transform, can get a clearer texture and richer detail fused image.