Image-Based Stamp Extraction for Enhanced Postal Automation

Postal automation is yet a partially solved challenge, and part of the challenge is the automatic and accurate extraction of stamps in envelope. We observed that the stamp perforations form a distinct boundary between the stamps and the background. The number and the linear distribution of stamp perforations can be utilized to extract and validate each stamp accurately. In this paper we approach this problem by proposing a novel stamp extraction and validation algorithm. First, the character of stamp boundary is emphasized and extracted through the RGB Components Absolute Difference under Variant Illumination Condition (CADVIC). Second, the stamp perforations are validated based on correlation matching, and then Hough Transform is exploited to generate the borderlines of each stamp. A number of experiments show our method is effective.