Study on weak bit in Vote Count and its application in k-Nearest Neighbors Algorithm

In the Vote Count for k-Nearest Neighbors (kNN) algorithm, the quantized projection values of query and training/reference vectors are compared and counted. In this process, not all quantized projection results are reliable for bit-matching and the search result may be distorted by these unreliable bit-matching. In this paper, the concept of weak bit is introduced to identify those unreliable bits after quantization and the corresponding bit-matching comparison is not executed. Simulation results show that, when weak bit is employed, the accuracy of kNN based on Vote Count can be improved significantly.