A Study Of The Watershed Algorithm On Multi-source Image Region Segmentation

Multi-source image segmentation which belongs to multi-source image processingtechnology has been an important field researched by scholars. Consequently, how touse the information of multi-source image to improve the image segmentation qualityand which effective segmentation method to be chosen has been a concern researchquestion. To solve this problem, this thesis studies the method of multi-source imageregion segmentation and focuses on a typical algorithm of the region segmentationalgorithm which is called watershed algorithm. This thesis puts to use watershedalgorithm to segment the multi-source image which mainly includes visible light imagesand infrared images.Compared with other methods, watershed algorithm can acquire continuous imageboundary, quickly compute and has a high calculation precision. However, watershedalgorithm has some disadvantages, for instance, it exists the over-segmentationphenomenon and is susceptible to noise. Therefore, in order to overcome thesedeficiencies of the watershed algorithm, this thesis mainly studies how to use regionsegmentation method to improve the watershed algorithm. The researched terms of thethesis are that:Firstly, this thesis studies the method of reducing noises, smoothing image andfiltering image which are applied to multi-source image. According to the principle ofmulti-source image and imaging characteristics, this thesis will put to use the bestmethod of reducing noises, smoothing image and filtering image to accomplish thepretreatment of the multi-source image, thus complete the improvement of the earlierstage of the watershed algorithm.Secondly, this thesis focuses on a region segmentation algorithm which is calledthreshold segmentation algorithm and carries on comparative analysis on the advantagesand disadvantages of threshold methods. This thesis selects the otsu method (OTSUmethod) as the threshold segmentation method, and further, the method is improved.This thesis proposes the otsu method combined with the gradient feature, and thus thismethod is adopted as the pre segmentation method before puting to use the watershed algorithm. Therefore this thesis also proposes watershed segmentation methodcombined with threshold algorithm, and further, the method is improved.Finally, this thesis deeply studies regional growth and division merge algorithmand according to the idea of the algorithm using clustering algorithms to solve the smallregion combining problem. This thesis carries on comparative analysis on theadvantages and disadvantages of the clustering segmentation method, and then selectsthe FCM clustering segmentation algorithm as the clustering segmentation method. Thisthesis proposes the FCM clustering segmentation algorithm combined with the spatialcharacteristics, and thus this method is adopted to improve the watershed algorithm.Therefore this thesis also proposes watershed segmentation method combined withimproved FCM clustering algorithm. Thus this thesis completes the ultimateimprovement measures of the watershed algorithm.