Low Contrast Line Segment Detection And Its Application

To obtain the lightning channel coordinates in digital lightning images and the morphology of retinal blood vessels,researchers mostly use manual processing method,but this method is not efficient and its result is often subjective.In order to improve the efficiency of data processing,realize the batching process,and rapidly analyze the features in the development and morphology of the lightning channel and the retinal blood vessels,more and more research is investigated to automatically recognize lightning channel information and retinal blood vessels information.Not all of the lightning channels are bright and clear.Some of the lightning channel images are complex and diverse because of low contrast,occlusion of clouds,and interference of other environmental factors such as buildings and raindrops.Therefore,a good lightning channel segmentation algorithm should not only has good effects for bright and clear lightning channels,but also has relatively accurate segment results for those channels with low contrast and complicated background.Accordingly,this paper put forward a new lightning channel segmentation algorithm based on line support regions.First,Gauss matched filter and contrast stretching method are applied to enhance the contrast of lightning channels,according to the gray distribution characteristics of cross section of lightning channels.Second,line support regions.which include lighting channels within a minimum enclosing rectangle,are extracted as foreground area by a line segment detection method.In addition,the line support regions are expanded in both the main direction and its perpendicular direction.Finally,Otsu thresholding method is applied in each line support region to extract lightning channels,since the gray level distribution of each line support region is bimodal.Furthermore,a new modified algorithm based on BLSR is put forward for lightning real time monitoring.First,LSD algorithm is applied to find the edges of the lightning channel.Second,SWT algorithm is applied to extract the stroke width at every pixel.Then,a width threshold is used to filter the pixels with large width.Finally,we judge whether a lightning flash occurs during a lightning process on the basis of the number of remaining pixel from previous step.For retinal blood vessel detection,a new method using double-scale nonlinear thresholding on vessel support regions for retinal image segmentation is proposed.First,two gauss matched filters with different scales are applied to pre-processing the retinal images and obtain two different MFR images.Second,for MFR image with the small scale,line support regions which include vessels within a minimum enclosing rectangle,are extracted as foreground area by a line segment detection method.Then,Otsu thresholding method is applied in large rectangle to extract retinal vessels,and Stroke Width transform method is applied in small rectangle to extract retinal vessels,since the gray level distribution of small rectangle is unimodal.For MFR image with the large scale,fixed ratio thresholding method is applied to segment the coarse vessels.Finally,we get the final segmentation result by fusion of fine and coarse vessels.