Multi-focus Image Fusion Based On Focusregion Detection

In optical imaging,the limited depth of field caused by the imaging principle limits the focus range,so it is difficult to avoid local image defocus blur completely.With the rapid development of information technology,image as one of the main forms of information,the images with local defocus blur cannot meet people's needs,so it is very important to deal with local focus or partial out-of-focus images.This paper is related to the study of images with local focus or out of focus.The specific research work is as follows:1 Non-reference image focus area detection based on BP neural networkThe existing learning-based focus area detection algorithm only pays attention to the characteristics of the original image itself,and cannot well represent the image focus sharpness or defocus blur.Aiming at this problem,this paper firstly explores the characteristics and differences of the image focus clear area and the defocus blur area according to the mechanism of image defocus blur.Considering that when the blur-filter is used to blur the locally focused source image,the quality degradation in the focused regions occurs significantly more than that of the defocused regions,and at the same time,in order to express focus sharpness and defocus blur from multiple aspects.Those feature,include the DCT,the singular value,the statistical difference and the edge energy,and those feature of the blurred version of the original image,are combined to form a feature vector which can distinguishes the focus region and the blur region.Then,when performing image sharpness measurement,the feature vector of the training image block is extracted and the training data set is formed for the feature vector mark definition,and the training data set is learned and trained by the BP neural network model to establish the image block clarity degree.Forecast model.Finally,the image to be measured is divided into several small blocks,and the image of the image block is measured by the degree of blur using the established network model to obtain a sharpness measurement map.Since the measurement results are fed back to the degree of clarity,in order to obtain the final focus area detection result,the focus area decision map is obtained by using threshold segmentation techniques,morphological operators and graph cut techniques.It can be seen from the results obtained from the simulation experiments that it is effective to detect the focal region using this method.Multi-focus image fusion based on BP neural networkThe existing multi-focus image fusion algorithms based on the focus region mostly adopt the method of multi-source image focus region fusion,and generally there is a problem that it is discontinuous between the clear regions or the sharpness of the focus region.Although the multi-focus image fusion algorithm based on multi-scale decomposition solves this problem to some extent,it often results in a decrease in the contrast of the focus area due to mis-selection of coefficients.In order to solve this problem,this paper firstly uses BP neural network to detect the focal region,and then uses the non-downsampled shear wave with good direction selectivity and shift selectivity to perform multi-scale geometric decomposition on the source image to achieve accurate mapping of the focal region.The non-focusing region is fused by using a combination of high-frequency coefficients and averaging of low-frequency coefficients.Finally,the fused image is obtained by inverse transforming the non-downsampled shear wave.It can be seen from the comparison experimental results that multi-focus image fusion using this method is effective.