Improved Non-negative Sparse Coding Image Classification Based On

With the development of internet technology, vast amount of images spring up. How to classify the massive image data into different categories based on image content is an important and meaningful task. Image classification techniques are proposed to deal with this problem. As a fundamental application direction in computer vision and machine learning. The Bag-of-visual-Words model and Spatial Pyramid Matching model which are extremely popular in image categorization have been widely used by many researches due to their very good performance. In the procedure of image classification, how to generate the visual dictionary is an important procedure.The traditional Bag-of-visual-Words model and Spatial Pyramid Matching model generally use K-means as the clustering method to form visual dictionary. However, because of the sensitivity to the initial center, the K-means algorithm is easy to converge to a local minimum, which seriously affect the formation of the visual dictionary. Sparse coding is a kind of signal processing method, which has been successfully used in the image classification task and achieved very good classification performance. Our innovative work is described as follows:(1) Implemented an image classification method based on sparse coding. In view of the advantages of the non-negative sparse coding method, we applied the algorithm to image classification task. Experimental results demonstrated the accuracy of our proposed method.(2) Proposed an image classification method based on laplacian non-negative sparse coding. Laplacian non-negative sparse coding can encode two features simultaneously and simulate the main visual cortex V1simple cell receptive field behavior of the mammal primary visual system with accuracy. Compared with the non-negative sparse coding method, the proposed method achieved better classification results.(3) Proposed a hypergraph laplacian non-negative sparse coding method for image classification. Hypergraph laplacian non-negative sparse coding method combined the hypergraph and the laplacian non-negative sparse coding to keep the similarity among the instances within the same hyperedge simultaneously, which improve the image coding ability significantly. Experimental results validated that the proposed method could achieve higher accuracy than other existing methods.