Research On Footwear Image Quality Assessment And Low-quality Footwear Image Classification Method Based On Video Surveillance

“Combining the footprints of the crime scene and the surrounding surveillance video to track the criminal suspect” has become a technique and method for the public security organs to detect various cases.On the one hand,the type of shoes can be inferred from the shoe prints left by criminals at the scene of the crime;on the other hand,the same type of shoes can be found in the surveillance video within a certain time and space when the crime occurs,and then the shoes can be used to find someone to track the criminal suspect.However,the process of identifying shoe types through video surveillance is relatively complex,which not only requires criminal technicians to have keen discrimination,but also needs to spend a lot of time and energy,and it is easy to miss the best time to solve cases.In real cases,the quality of video surveillance is not high,which increases the task difficulty of criminal technicians.In order to solve this problem,an image quality assessment method for footwear images under video monitoring is firstly proposed,which is used to screen high-quality and low-quality footwear images.Then an automatic classification method of low quality footwear is proposed for rough classification of images of low quality footwear that are difficult to identify specific shoe types.The work of this paper includes: according to the shoemaking standard formulated by the national shoemaking standardization technical committee in 2017,shoes are divided into two categories: leather shoes and sports-casual shoes(include sports shoes and casual shoes).Two databases of footwear image quality assessment and low-quality footwear image classification were constructed.The former included 54 630 high-quality footwear images and 46 973 lowquality footwear images,while the latter included 66 425 leather shoes images and 109 671sports-casual shoes images.Based on the convolutional neural network,the network model of footwear image quality assessment and low-quality footwear image automatic classification are designed.The experimental results show that the network model of footwear image quality assessment has good performance in the aspects of brightness assessment,contrast assessment,definition assessment and background interference assessment.The automatic classification model of low-quality footwear image has been verified by fivefold cross validation,and the accuracy has reached 98.5%.It can be seen that the designed network model can accurately identify the images of low-quality and high-quality footwear,low-quality leather shoes images and low-quality sports-casual shoes images.