Segmentation Of Flat Shoeprint Image Based On Semantics

In the field of forensic science,the flat shoeprint is a kind of important mark evidence,which can indirectly provide information such as the age,height and gender of the suspect.With the development of computer technology,Automatic identification of shoeprints becomes an important part of modern shoeprint identification.The existing shoeprint analysis focuses on extracting the sole pattern and establishing the pattern database,and there are few researches on the image features that have strong identification significance in contour and sole wear.We systematically segment the pattern,sole wear and contour of the shoeprint,which reduces the subjectivity caused by manual labeling in shoeprint identification and lays the foundation for shoeprint identification.Image feature segmentation based on semantics on flat shoeprint refers to the interpretability of "generalized" semantics-making the segmented class results meaningful in the field of shoeprint identification.In this paper,more than 7600 flat shoeprints printed by ink are used as research samples,and different image segmentation methods are applied to realize the segmentation of three types of flat shoeprints.In terms of segmenting the shoeprint image to obtain the pattern features,the paper proposes to segment using local intensity clustering and multiplicative intricative component optimization(MICO),with the goal of obtaining a highly visibility of sole pattern.The results show that the local intensity clustering method is robust to pattern segmentation.The MICO algorithm makes the offset field of the shoeprint image corrected,the image becomes orderly and less noise,and the pattern is effectively segmented.The scores of 600 segmentation renderings are evaluated,and the success rate of pattern segmentation is 83%,and the performance is excellent.In terms of segmenting the shoeprint image to obtain the sole wear features,the pseudo-color and mean shift methods are proposed to segment the pre-processed image by the MICO algorithm to obtain a "colored" flat shoeprint wear zone.The results show that the pseudo-color improves the recognition of the wear zone of the shoeprint,but it sacrifices the accuracy of the shoeprint image to a certain extent,and the effect is good.Mean shift can effectively segment the wear zone without prior knowledge,and it has strong robustness and excellent performance.In the evaluation stage,the effect image features obtained by the mean shift division are matched,and the results are good.In terms of segmenting the shoeprint image to obtain the contour features,the shoeprint is segmented using three methods: edge operator,threshold and region segmentation.The aim is to enhance the discrimination between the shoeprint area and the non-shoeprint area on the image.The results show that the edge operator has a general effect on contour detection,and the adaptive threshold method performs well.The performance of segmentation contour is better in the region growth method,and the watershed method is over-segmented and performs poorly.The contours obtained by the segmentation are evaluated by the relationship between the height and the forefoot width,and the results are consistent with the facts,which proves that the contour segmentation is effective.