Adaptive weighting of local classifiers by particle filters for robust tracking
Pattern Recognition
Local normalized linear summation kernel for fast and robust recognition
Pattern Recognition
Hi-index | 0.00 |
This paper presents a local normalized linear summation kernel. Kernel based methods are effective for object detection and recognition. However, the computational cost is high when we use kernel functions except for linear kernel. To realize fast and robust recognition, we define normalized linear kernel. The normalized linear kernels are applied to local regions of a recognition target, and the kernel outputs are integrated by summation. This kernel is called as "local normalized linear summation kernel". We show that kernel based methods with local normalized linear summation kernel can be computed by linear kernel of local normalized features. Thus, the computational cost of the kernel is nearly same as linear kernel and is much lower than RBF or polynomial kernels. Effectiveness of the proposed kernel is evaluated in face detection and recognition problem. We confirmed that our kernel gives better accuracy with low computational cost than RBF and polynomial kernels. In addition, our kernel is also robust to partial occlusion and shadows on faces because it is based on the summation of local kernels.