Fast Logo Detection In Scene Images Based On Histogram Of Sparse Coding

Logo is a symbol of an enterprise which is different from other enterprises, and the Logo plays a unique role in enterprise and also plays a role of promotion. And the technology of Logo detection detects and recognizes the Logo of an object with use of image processing technology. Logo detection has a good prospect in the field of enterprise brand tracking and analysizing of market share, for example, an enterprise can get the pictures of micro blog and track the brand with more applications. Although Logo detection has gained improvement in the past few years and some scholars also obtained some research results, the speed of detection still has more space for improvement.In this paper, an improved method for the detection of the candidate region is adopted. This method is based on the suppo rt vector machine classification and gradient feature of machine learning to realize the candidate location of the Logo. The Logo is attached to the object, so if you directly find a relatively small Logo in a picture,traditional methods may lead to an increase in the complexity of the algorithm. Method with the rapid of locating objects will speed up the calculation.We adopt the method of the p ixel feature extraction based on the histogram of sparse encoding. In the process of using sparse representatio n of pixel features, we use K-SVD for dictionary learning. Compared with the traditional method, the combination of sparse representation and histogram can express more detailed message.In view of the advantages of the above method s, a new algorithm is proposed to combine the two methods which are improved generic objectness and histogram of sparse coding. The images were scaled to the proportion of different combinations of the collection, then the gradient feature of the image is used to classify, and Logo candidate regions were screened. Then we select the frame of which the scores are greater than the threshold of the border as a collection of fine classification and detection. And then use the histogram of sparse coding to detect and classify in the set of candidation. Adoption of the method of the gradient plus-bit computing can improve the speed, and the sparse coding can reduce the image noise and provide more image information so as to improve classification accuracy.In order to verify the method p roposed in this paper, the speed testing experiment and the precision test experiment are carried out respectively. On the premise of this paper, through the test experiment, it is verified that the combination of the two methods proposed in this paper has some improvement in speed and accuracy.