Research On Interactive Scene Shoeprint Extraction Algorithm

The shoeprint is one of the important physical evidences in crime scenes.Accurate extraction of shoeprint in the environment of complex crime scene is one of the urgent problems to be solved in footprint recognition technology.The key job of extracting shoeprints is to separate the pattern area from the complex background.Different from other types of images,the background of shoeprint images is complicated.It is difficult to obtain the ideal image segmentation results by traditional image segmentation methods.The main purpose of this thesis is to obtain accurate and noise-free shoeprint images from the complex background.Interactive image segmentation acquires results by means of human-computer interaction.Firstly,the user makes the target and background markers on the image,and these markers can be used as a priori information to guide the segmentation.Then the segmentation algorithm model is established according to the priori information.Finally,the segmentation result is obtained by the algorithmic model we built.Based on the purpose,the thesis presents the algorithm of interactive scene shoe print extraction.The main works are as follows:1)The overall idea of an interactive scene shoeprint extraction algorithm is proposed.In connection with the shortcomings of the current shoe pattern extraction algorithm,a general framework of the interactive scene shoe pattern extraction algorithm is given.The framework mainly includes three parts:the region merging algorithm considering the directional features of a scene shoe pattern,the kNN-based interactive image segmentation optimization algorithm and multi-scale morphology based image enhancement algorithm.2)The regional merging algorithm considering directional characteristics is proposed.Based on the analysis of the characteristics of scene shoeprints,this algorithm improves the traditional regional merging algorithm.According to the similarity measure method of directional field feature and color feature,the similarity of adjacent regions is obtained,and combines the semantic information acquired by interactive,the regional merging method which conforms to the scene shoe print image is proposed.3)An interactive image segmentation correction algorithm based on kNN is proposedIn view of the limitations of traditional interactive image segmentation algorithms,the process of extracting shoe prints is gradually carried out.When the results of the pattern extraction are not satisfied,the kNN classification strategy is used to correct the pattern extraction results until the satisfactory results are obtained.The experimental results show that this algorithm can reduce the background interference and keep the details of shoeprints,and improve the accuracy of extraction of shoeprints.4)An image enhancement algorithm based on multi-scale morphology is proposed.In order to obtain an accurate and clear shoe print image,the thesis proposes an image enhancement algorithm based on multi-scale morphology.The algorithm overcomes the limitations of the traditional multi-scale Top-hat algorithm and it is robust to illumination variation and obtains the clear shoeprint image without background interferences.In order to verify the performance of interactive scene shoeprint extraction algorithm,we construct a dataset composed of 100 scene shoeprint images randomly selected from 1189 crime scene shoe images.We use four evaluation indexes to determine the accuracy of segmentation results.The experimental results show that the proposed algorithm is an effective method.