Evaluation Of Illumination-Robust Feature Point Detectors

Feature point detection is a basic problem in image processing.The accuracy and eficiency of feature point detection will directly affect the performance of subsequent image process-ing.Feature points,as an important feature of images,are widely used in image matching,pattern recognition,and other fields.Because there are many advantages of feature points compared to other locally invariant features in images,and when using feature points,only a small amount of information is needed to describe the main information of the gray change in images.Therefore,studying feature point detection algorithms have an important signif-icance.However,the performance of feature point detection algorithm is difficult to be simply de-fined as good or bad.There are many criteria for evaluating the performance of feature point detection algorithms,such as repeatability for judging adaptability,detection time for judging efficiency.Different criteria evaluate the performance of the algorithms in different aspects,the obtained results usually have ambiguity.Therefore,the comprehensive evalua-tion method in fuzzy theory can be used to evaluate the performance of the image detection algorithms.By using fuzzy evaluation judgment method,a mathematical model of evaluat-ing feature point detection algorithms synthetically can be established based on experimental data.Then a more scientific evaluation conclusion can be obtained after indicating the degree of which the assessment results are subject to.The main work of this thesis is as follows:(1)This thesis selected seven feature point detection algorithms that are widely be used,summarized the principle of each algorithm and reproduced them by MATLAB.Then the above algorithms were applied to multiple sets of images with illumination variation,and recorded the evaluation data of the following three aspects:number of feature points detected per image for judging quantity;repeatability per image for judging adaptability,detection time per feature point for judging efficiency.(2)By combining the above three evaluation criteria of the performance of different algo-rithms,this thesis focuses on the establishment of fuzzy comprehensive evaluation model and the determination method of weight.After a set of evaluation experiments,a better evaluation result was obtained compared with the former single criterion evaluation system,verified the practicability and validity of the evaluation model.