The Research Of Object Recognition Based On Feature Fusion

Object recognition technology is an important computer vision research; it has been widely applied to practical life. Object recognition is now mainly in the generic object recognition or object classification. One of the key problems is to find an effective feature can represent or distinguish the objects. Therefore, in-depth researches of object feature extraction help to promote the maturity and development of object recognition technology. It is of great research value and broad application prospect.In this paper, the global and local features of object are researched, and the existing methods of features extraction are introduced and analyzed. This paper also put forward a new object classification algorithm which combines global and local features, and the experimental results show the effectiveness of the algorithm. The specific work and results are as follows:1. The research of image segmentation and similarity matching based on global color feature is represented. Aiming to improve the segmentation speed while maintaining good segmentation results, a GrabCut image segmentation algorithm based on wavelet transform is proposed in the first place. Then, this paper proposes a method of the color histogram similarity calculation in the HSV space model, and compares it to the color histogram in the RGB space model. The results of the comparison shows that the similarity measure based on HSV color histogram is more effective and the limitations of the global color feature.2. On the basis of local invariant feature (SIFT), this paper propose a method that using principal component analysis (PCA) to reduce the dimensionality of the feature vector and improve the speed of feature description and matching. After the rough matching based on PCA-SIFT, it uses the random sample consensus (RANSAC) algorithm to remove the wrong matching points. Finally, experimental results show the effectiveness of the proposed method.3. According to the complementary of global and local features, a new object recognition algorithm based on combining global and local features is proposed. For global features, Hu’s Moment Invariant and HSV color histogram are computed for the image, and for the local features SIFT descriptor is used. In this framework, global and local features are extracted in the first place. Then Support Vector Machine (SVM) is applied to classify the test images and three different classifications are represented. Finally, the algorithm uses a fusion rule (there are rules of addition, multiplication, voting, etc) for decision fusion of classification results and obtains the final classification result. The experimental results show that the algorithm can effectively improve the accuracy of the object classification or recognition, and the algorithm for optimal while using the fusion rule of addition.