ISNN'10 Proceedings of the 7th international conference on Advances in Neural Networks - Volume Part II
Global matching to enhance the strength of local intensity order pattern feature descriptor
ISNN'13 Proceedings of the 10th international conference on Advances in Neural Networks - Volume Part I
| Hi-index | 0.00 |
Semantic image annotation can be viewed as a classification problem, which maps image features to semantic labels, through the procedures of image modeling and image-semantic mapping. In order to improve the performance of image modeling, we propose a novel method which is based on affinity propagation (AP) algorithm. For a given image, low-level image features are extracted from image sub-blocks, and the image feature distribution can be modeled by a mixture of Gaussian components. An adaptive mixture component number selection algorithm which is related to the image semantic information is also developed. The AP algorithm is adopted to improve the efficiency and accuracy of the distribution estimation. For a given label, the overall distribution is modeled, and the mixture component number is selected according to the mixture exemplars extracted from all images and the average value of the preference parameter. The experiment results illustrate that the proposed algorithm has the higher efficiency and accuracy compared with C-means and expectation-maximization (EM) algorithm combination.