Research On Local Blurred Image Detection Based On SVM

Images become one of the most important sources of information dissemination,they play a significant role in communication.The most commonly tool for recording image information is the camera.However,the images inevitably exist the phenomenon of image degradation,such as blurred,color cast.And the images quality directly affect the result of communication.In this paper,we analyze the forming mechanism of local blurred images at first,including motion blurred and out-of-focus blurred images.Comparing with the degradation model of two local blurred images,we can find the difference between the two.And,we also can recognize the difference between clear images and blurred images.We propose a locally blurred image classification for recognizing the blur types without needing to perform blur kernel estimation.Because it is hard to truly estimate blur kernel of an image.In order to better classify the blurred images,we extract four image features as characteristic vector.Extracting multiple image features can better represent the characteristics of images from various aspects.On the database,the SVM adopt all features of images to train them and generate strong classifier.The SVM can classify images into three types,clear,motion blurred and out-of-focus blurred.Finally,we detect the blur information of one pixel by using a local patch centered at it.Due to the misjudgment of SVM classifier,the final detection images contain some small empty or unconnected regions.Then,we used the mathematical morphology algorithm to perfect final testing results.The classification experiments showed that those blurred features could separate blurred images form clear images exactly,and could identify the blurred types.The high accuracy rate of classification indicated that this model is feasible for local blurred images detection.And the detection experiments on the database showed that local blurred image detection based on SVM has achieved a satisfactory result.Compared with the ideal detection images,the algorithm can also accurately detect the blurred regions in a local blurred image.