Infrared Image Threshold Segmentation Research Based On Otsu

Being the basis of feature extraction and recognition of image analysis andunderstanding, image segmentation is the technology to segment out the region ofimage that people interested according to a certain rules. After segmentation, the sameregions have the similarity and the different regions have differences. Because ofsimplicity in principle、less computation and good segmentation effect, Otsu, one ofthe threshold segmentation method, is widely applied.Infrared image is so a special class of images that it has the characteristic of lowresolution and low signal-to-noise ratio. It is of great importance in terms of bothmilitary and civilian to study on infrared image segmentation. This paper introducesmethods of infrared image segmentation and improvement base on Otsu. The mainpoints are as follows.Firstly, on account of the introduction of within class variance, it is effective toimprove the segmentation results when using the maximum scatter difference. But,the better result is on condition that the parameter C is adjusted artificially many times.So, a1-D improved maximum scatter method is proposed, which can segmentinfrared images adaptively and get a better segmentation result.Secondly, one2-D threshold average method and one improved Otsu methodunder the constraints of the range which is combined with LBP are put forward.Threshold average method bonded methods of maximum variance between clusterand minimum variance within cluster that can improve segmentation result. And, withregard to the improved Otsu method combined with LBP, a new algorithm is givenout to gain neighborhood average according to LBP that can get better noiseresistance. Meanwhile, there is obvious peak or trough information in2-D histogram.So, it’s able to enhance the effect of segmentation better.Then, one2-D gradient method of generalized probability with whichmorphological transformation is introduced to combine is proposed. By means ofmorphological preprocessing, the characteristic of target and background is moreobvious. What’s more, it is more accurate to regional division on2-D gradienthistogram, because the size information of gray-level is added into the histogram.This method can improve the infrared image segmentation result much effectively.Finally, artificial fish-swarm that is a simulation of animal behavior has the good ability of obtaining global extreme and to avoid falling into local extreme. It is a quickand efficient optimization algorithm. Furthermore, it is quicker to realize the goal ofoptimization when adding artificial fish-swarm algorithm into segmentation andchanging iterative step.