Research And Implementation Of Envelope Image Retrieval System

With the rapid development of science and technology, more and more communication means have become available for people, apart from traditional mails. However, there are still a lot of traditional mails circulating in the world every day. During the procedure of mail processing, it is usually referred to the information query of mail, which is realized by barcode through manually keying in the barcode or automatically recognizing the barcode. It costs a huge manpower and material resources, reducing the competitive capability of postal enterprises.Considering that one mail may pass through more than one sorting machine, or may pass through a sorting machine more than one time, and its image could be captured during the procedure of mail processing, it is possible to realize the query of mail information on the basis of envelope image retrieval.The paper takes envelope image with handwritten address characters as object, analyzes the characteristics of the images, and selects appropriate image features, and then designs a feasible and effective flow for envelope image retrieval.The paper employs partition-block image feature, projection histogram feature and SIFT feature to construct a feature vector, and designs a serial flow for operation to realize effective retrieval. It is demonstrated by experimental results that our retrieval system can reach at least 98% accuracy, which is meaningful for practical applications.