Research On Detection And Segmentation Algorithms Of Local Blurred Images

Image is the carrier of information.In practical applications,the information conveyed by image is of great significance in most fields.When the image is captured,the optical image is focused by a lens.The local blurred image is caused by the difference in the depth of field of imaging areas or the position change of the focus area in the imaging system,which greatly affects the acquisition of critical information.Therefore,it is necessary to enhance the local blurred area of the image,which requires the effective detection and segmentation to the blurred image area.This thesis mainly focuses on the detection and segmentation algorithms of local blurred images,and the enhancement of the blurred regions of partially blurred images.Main work is as follows:(1)According to singular value decomposition,the blurred intensity of sub-region of the image is calculated and evaluated.The natural blurred images of different types can be simulated by blurring model from clear image.It is necessary to study the formation of different blur types and the derivation of blurred model.According to formation methods,image blur is generally divided into defocus blur and motion blur.The defocus spot on the image plane when the object is not in the focal plane of the imaging system will cause the image to be out of focus blur.The movement of the object or camera shake will cause the same object to be imaged at different positions on the image surface during the exposure time,which will cause motion blur.Researching on two types of blur causes and analyzing the difference between the blurred area and the clear area after re-blurring through the re-blurring theory,and using the characteristics of singular value decomposition increasing the difference between the image blur and the clear area before and after re-blurring processing,to obtain the local blur detection index-blur intensity.(2)According to the re-blurring theory and the definition of image blurred intensity,partial blurred image is divided into blocks at a certain scale,and the blurred area in the image is located by the sub-block blurred intensity.A multi-scale block fusion detection algorithm is proposed to position the local blurred area from the large-scale block roughly.On the basis of the rough positioning,combining with the fuzzy positioning under the small-scale block,the blurred image region is detected and segmentation accurately.The K-means clustering algorithm is used to classify the blurred intensity of the sub-image adaptively,so as to realize automatic detection and segmentation of the image partial blurred regions.(3)Research on the enhancement method of local blurred image based on the detection and segmentation of blurred area.According to the image degradation model,an improved Wiener filtering algorithm is used for image enhancement,and the information of partially blurred images is recovered.Detection and segmentation of local defocus or motion blurred image and enhancement to blurred regions are experimented.The reliability of the algorithm in this thesis is verified.Compared with corresponding algorithms,the results show that multi-scale fusion segmentation algorithm presented in this paper has the characteristics of small computation and high segmentation accuracy,which is a universal algorithm for detecting and segmentation of single image and local image,and can be applied in most areas.