Research On Image Segmentation Evaluation Based On Feature Extraction

Image segmentation is a prerequisite for image processing and is widely used in the field of image analysis.There are many image segmentation algorithms,and segmentation results are divided into good or bad quality.Therefore,researchers have proposed many methods for evaluating the results of image segmentation.How to effectively measure and appropriately select segmentation evaluation method is very difficult.In order to study this problem,this thesis mainly does the following work:(1)From the perspective of application,segmentation evaluation methods are classified,commonly used supervised and unsupervised evaluation methods are summarized.Common methods of supervised and unsupervised evaluation are compared and analyzed on natural,medical and remote sensing images,and the effectiveness of these evaluation methods is ranked.By summarizing research status and experimental comparison and analysis,it provides a basis for the selection of different types of image segmentation evaluation methods.(2)This thesis proposes a new evaluation method based on edge detection and feature extraction to evaluate the quality of image segmentation.This method belongs to unsupervised evaluation.This method is simple and easy to implement,and the detection and extraction methods can be changed according to your needs.This thesis proves effectiveness of the proposed method on natural,medical and remote sensing images.Experiments of natural images are carried out on four different datasets of Berkeley,Weizmann,Simfukwe's and VOC2012,and the accuracy of 98.67%,99.00%,98.50% and 97.50% is obtained respectively.The method is compared with some classic and newer evaluation methods,and it is found that its performance is better than most existing methods.