Quality Pre-detection System For Identification Photos

With the advance of e-government system, online identification photo making enters into public view. Typically, the pipeline of identification photo processing includes photo shot, image cropping, color optimization and background replacement. Artifacts such as clutter background, fuzzy, noise and unqualified exposure in photos took by users will bring a lot of trouble or even result in failure to the identification making. Hence, it is necessary to detect the image quality and ask for the user resubmitting a new photo if the detection does not pass the quality detectionThis paper proposes a quality pre-detection system for identification photos which includes background classification, fuzzy measurement and exposure evaluation. The photo breaks one of the three conditions will be judged unqualified..we proposed an image segmentation algorithm based on superpixel clustering to partition the photo into foreground and background regions.. We then employ HOG+SVM to classify the background region into simple type and complex type. We employ a re-blurring method to perform fuzzy measurement.. It first detects the edge pixels in the image, and then calculates a gray value variation for the area of edge pixels. This finally leads to a fuzzy index of the identification photo. Exposure detection includes two steps. It first employs machine learning to detect global underexposure and overexposure artifacts and the uses symmetrical principle to detect the uneven exposure issue.The main contributions of this paper is as follows:(1) It proposes an image segmentation algorithm based on superpixel. To make the algorithm robust and efficient, we cluster the photo into many superpixels using the SLIC algorithm. Then, we create an undirected weighted graph for the superpixel image and segment it using the Grabcut method. Considering that the segmentation accuracy of the Grabcut is largely dependent on the input mask as priors, we make use of a set of algorithms to detect the hair region and the skin region as foreground or probable foreground regions and then have improved the accuracy of segmentation and identification photo matting;(2) it designs a quality pre-detection system of identification photos which includes background classification, fuzzy measurement and exposure evaluation. You can filter out unqualified photos quickly and effectively using this system.