Barcode detection and recognition using Wavelet transform

Abstract

Wavelet! Ondelettes! These words seems nowadays to be brought into fashion. Still there is no question of poetry through these synonyms. Would they become the stock reply answers for signal processing? Originally developed in the eighties for a geological oil layers localisation, the theory get applications in signal compression, data visualisation, pattern recognition, etc. The appealing algorithm both rapid in o (N) and apparently self evident with its filter banks structure made of vulgar high and low passes, attracts sympathy. The JPEG 2000, appealed to replace the old JPEG uses the DCT, is based on the Wavelet Transform. It allows a good maintenance of the picture quality with artefacts less cumbersome than the blocking effects. In the same way for Pattern Classification and Recognition, the Wavelet Transform let decompose the time (or location)-frequency space in an advantageous way by optimising the vector basis selection. Consequently, the idea of the research was to make use of the wavelet theory for a joint Picture Compression and Pattern Recognition. For this purpose, the problem was cut into two sub-projects: Compression and Pattern Recognition. This present document will deal with the second part. The subject was expressed by the following words of mouth. A sequence of pictures coming from an autonomous robot equipped with a low resolution camera should be compressed and then analysed in real time. Within the shouted scene should be located and decoded a simple barcode from free election standard. Of course, the robot was an application example, which was not aimed to be concretised. The classification problem may then be divided into the stage of feat ure extraction, dimensionally reduction and pattern recognition. As a matter of fact, due to the high dimension of linear picture-domain-frequency representation, its success lies upon an appropriate form of dimensionally reduction.