Extraction Algorithm For Shoeprint In Complex Background

As one of the traces of crime scene,shoeprint have a high reference value in criminal investigation.Accurate extraction of shoeprint in the complex environment of crime scene is the basis and key of subsequent shoeprint retrieval tasks.The main purpose of this thesis is to process the shoeprint images collected from crime scene and extract accurate shoeprint images.Based on this purpose,this thesis studies the extraction algorithm of shoeprint under complex background.:(1)The general idea of the shoeprint pattern extraction algorithm under complex background is proposed.Aiming at the shortcomings of the existing shoeprint extraction algorithm,the overall framework of the shoeprint extraction algorithm incomplex background is proposed.The framework includes:regional feature extraction algorithm for shoeprint image,shoeprint candidate region extraction algorithm based on random walk,and shoeprint image post-processing algorithm.(2)A region feature extraction algorithm for shoeprint images is proposed.In order to extract stable pattern region features,a region feature extraction algorithm for shoeprint image is presented in view of the diversity of pattern bearing scenarios in complex background.Firstly,the shoeprint image is superpixel divided.Because the pattern has the characteristics of hollow discontinuity,there will be a small amount of background residual in the super pixel area containing the pattern.In order to reduce the interference of redundant information,the characteristics of the pattern in the area are accurately deseribed.The local adaptive gradient threshold model is used to extract the gradient features in the region,and the gradient and color information are used to describe the regional features.(3)An algorithm for extracting candidate regions based on random walks is proposed.Through the analysis of the characteristics of the shoeprint image in complex background,it can be seen that the background of the image is diverse and uneven,which leads to the unsatisfactory effect of extracting the pattern directly from the original image.Therefore,the process of pattern extraction is carried out hierarchically in this thesis.Firstly,candidate regions are extracted to remove most of the background noise interference.Remove most of the background noise interference.In this thesis,the classification of random walk is improved.Combining the idea of weighted K-nearest neighbor classification,the accuracy of regional classification results by random walk is improved,and the effect of extracting candidate regions of pattern is improved.(4)A post-processing algorithm for shoeprint image is proposed.In complex background,the contrast of the whole pattern is not uniform.Some image patterns are shadowed,resulting in the local pattern is not obvious,and there is residual background noise between the hollow pattems.Therefore,it is necessary to enhance the image of the extracted candidate areas of pattern,and to enlarge the contrast between pattern and residual background.This thesis combines morphological gradient information and multi-scale top-hat transform to realize adaptive image enhancement.This method can adaptively enhance the contrast between the pattern and the surrounding background,and suppress the amplification of noise,so as to keep the overall enhancement effect of shoeprint uniform.In this thesis,100 shoeprint images are randomly selected from the complex background data set provided by public security organs as test data,and three indicators are used to determine performance.The experimental result show that the proposed algorithm is an effective algorithm for shoeprint extraction.