Object Identification Algorithm And Application Research Based On Hyper-feature Patches

Identification is become more and more popular in the modern world. The biological characteristic identification become the main technology because of its higher safety, credibility and effectiveness. Biological characteristic identification technique use body biological characteristic to personal identification. Because of its advantages of the user less participation, data collection of no contact, no injury to the user and easily conceal, Face identification technique as one of the biological characteristic identification techniques is easy for the user to accept, which is called the most promising identification method in the 21st centuries. Therefore , face identification technique has broad application prospect and become one of the hot studying topics in the field of computer vision and pattern recognition.However, face identification techniques always face a challenge when the training samples are few or even just one training sample , the recognition rate of the present popular face identification algorithms is extremely low.This paper analyzed and studied the popular face recognition algorithms which have their own applicable environment, advantages and disadvantages. However, their common shortcoming is the recognition rate is low when training samples are few. The paper studied and implemented the face identification algorithm based on the hyper-feature patches and the algorithm can settle theproblem----few training samples caused low recognition rate. Meanwhile,compared to the other algorithm, it was superior in training speed. The basic idea of the face identification algorithm based on hyper-feature patches is to divide the object face image and the test face image into patches , compute distance between object-face image patch and test-face image patch by matching each patch , then make use of the simple Baysian model to find matching score of the two face image by the probability statistics of the distance, finally, make the decision by the score . The paper also studied and implemented the PCA algorithm based on Baysian classification .The method once acquired the champion of face identification in 1996. But when training samples are few, its recognition rate is also low.Considering the characteristics and advantages of the hyper-feature patches technique and PCA technique, this paper proposed a fusion method of this two techniques. The traditional method of patch matching is to match pixel by pixel , thus the computational quantity is too huge to meet the real-time requirement. Whereas, the algorithm proposed in this paper adopted hyper-features which consisted of few features to express the image patch and find patches' s match by hyper-features. Therefore, the selection of the hyper-features directly affect the recognition rate of the algorithm. Based on the orthogonalization principle, the PCA technique is used to reduce dimension and orthogonalize. The fusion of techniques which was proposed in the paper was introduced as follows. Firstly, the face image was divided into patches. Secondly, PCA technique was adopted to reduce dimension and orthogonalize on the image patches. Irrelevant features value which was got from the reduced-dimension and orthogonalization patch is regarded as the value of the hyper-features. Finally, the hyper-features which were gained from this method represented the image patch and matched the patches. These features in the hyper-feature were irrelevant , so they can more effectively represent the image patches. The experimental result shows the fusion technique is more effective.