Research On Image Local Blur Detection And Segmentation

People often use cameras to record related information in life and get a lot of picture resources.Due to the constraints of human factors and environmental factors,the phenomena of blurring often appears in the process of image acquisition.Image blur will weakenthe visual effect,resulting in a decline in the ability of information transmission.The image blur interferes with resolving ability of eyes,detecting and segmenting blurred region of local blur image can be applied toimagerestoration,object recognition,imagefusion andso on.This thesis studies the formation mechanism of local blur images,compares the mechanism of defocus blur and motion blur,analyzes the difference between blurred image and clear image,and points out the theoretical feasibility and application value of local blur image detection.The shortage of the current image blur detection: the texture smooth region and the color uniform region are difficult to be correctly divided,partial blur detection is only for one of motion blur or defocus blur.Aiming at the shortcomings of partial blur detection methods,a local blur image detection algorithm based on image block re-blur and correlation coefficient are proposed based on previous works.In order to better reflect the blur character,the improved covariance matrix similarity and the singular value decomposition method are used to calculate the correlation coefficient of the image block,We set the fixed threshold to complete the mark of the blur image block.The experimental results of partially motion blur and defocus blur images show that the method is feasible,accurately judges the blurred degree of the image block and can detect some smooth regions of texture.A single blur feature is not stable to distinguish the blur region,the current blur detection method is difficult to segment the local blur image accurately.In order to distinguish the blur region,a local blur image segmentation method based on pixon is proposed,which combines the gradient histogram span,the local binary patterns and the kurtosis.Then constructs the image block with the pixel point as the center,three characteristic value of the image block are extracted as pixel blur degree,and the pixons are constructed by blur map.Finally,the fuzzy C means clustering is carried out according to the given blur degree of pixel,and pixon classification is completed.The final experimental results show that our method can apply to partially motion blur and defocus blur images,segment blur region more accurately than really ground truth,and it can effectively distinguish part of texture smoothing area,has a certain resistance performance to noise,represents blur features from many aspects.Our method also has some good effect in the mixed blurimagedetection and theblurmap is smooth and continuous.