Automatic Translation System Of English Postal Mails Based On Image Recognition And Address Translation

Machine translation is a critical and important subject in natural language processing. It has a broad market potential and research value in the rapid development of information society. With the development of science and technology, machine translation based on image recognition begins to attract people's attention and concern. The translation of the international postal mails is a process of translating the English address into its corresponding Chinese, and marking on the postal mail for further sorting and delivery.This paper is based on the project of SATP-120 high speed translation and sorting machine, studying the translation process of the images for international postal mails The paper introduces the process of machine translation technology and its related research, presents the technology of location and recognition of the barcode, and proposes the automatic translation system based on image recognition and address translation. The proposed system is composed of the following modules: barcode location module, barcode recognition module, addressee address location module, OCR recognition module, address recognition module and address translation module. The main jobs of this paper are as folIows:First, the image of English addressee address is automatically located based on image processing algorithms, and converted to text information by OCR method. The OCR character post-processing based on address lexicon is used to decrease the interference of error OCR characters.Second, the paper employs the finite automata, to automatically recognize the address information, such as names, road names, numbers, etc. What's more, number sorting rule is designed for the current demand;Third, a translation strategy based on a combination of address example database and literal translation is proposed and demonstrated with good translation effect.Fourth, a barcode location approach is proposed, according to the gradient features and corner features of the barcode. The approach can automatically distinguish barcode from other similar textures, such as the text and so on. Besides, an improved Bernsen binarization algorithm is proposed which is capable of removing the impact of uneven illumination effectively; The paper validates the system's performance by experiments. As is shown in the experiments, the ideas and approaches in this paper are effective and feasible The research and its outcome have some reference values and good application prospect in the postal automation field and other related areas.