Research On Logo Detection And Recognition Based On Convolutional Neural Networks

With the development of Computer Vision, object detection technology has been utilized more and more widely in real life. For example, logo detection and recognition has been proved to be an efficient solution to dealing with massive commodity images in e-commerce. First of all, by detecting and recognizing the logo in product images, the brand can be identified and can be judged from fake goods with further analyzing the texture information. In addition, logo detection and recognition can be the basis of image-based and brand-related searching. For these reasons, this paper proposes a method for logo detection and recognition based on convolutional neural network.Object location is another key difficulty in object detection task besides identifying the target category. We solve this problem in three steps. First is to design a binary classification neural network to distinguish between logo patch and non-logo patch. Then, both positive and negative training data are generating from processing the real samples. At last, the weights and bias are learnt through back-propagation algorithm in the training dataset.Given the logo position, we use a convolutional neural network based recognition method to further decide the category. Firstly, we adjust the number of feature maps, the number of the fully connected layer nodes, and other parameters according to logo recognition target. Secondly, the detected logo regions are utilized to learn the network parameters.Experiments demonstrate that the proposed method can handle the multiscale, rotation and deformation problems and achieve well detection results.