2-Dimensional Marked Object Localization In Complex Background

2-dimensional bar code is very useful in many important applications. Because of the various appearance and complex background, there has not been any helpful localization algorithm which can be applied to various materials (such as plastic,metal,paper). The thesis represents a novel algorithm to locate 2-dimensinal bar code, which is based on machine learning. First, the algorithm divides the input image into sub-blocks, and extracts a feature vector which can describe the texture characteristic for each sub-block. Then the algorithm takes the feature vector as input, classifies the sub-block with Adaptive Spatialboost. It tells whether the sub-block belongs to the 2-dimensional bar code or the background. Finally, after the post process, we get the localization result. We propose the Adaptive Spatialboost on the base of Spatialboost. It can combine the sub-block's texture information with the connection between sub-blocks adaptively. So the classification result of one single sub-block is not only decided by its own texture feature, but also relies on the labels of its neighborhood. This can help it to get more precise result. We evaluate the algorithm on a testing set, and the testing result shows a satisfactory performance of the method.