A new local binary pattern for smoke recognition

Abstract

In order to improve the detection rate of smoke recognition and reduce the false positive and error rates, a new local binary pattern (Zigzag Local Binary Pattern, ZLBP) is proposed. In ZLBP, we first rearrange the pixels in the local area into four linear areas by four zigzags with four directions, and then design two coding methods for the linear areas. For four linear areas, we can get four feature vectors, each of which is computed based on two codes of the same linear area. Finally, we concatenate the four feature vectors to generate ZLBP feature. Experimental results show that the new proposed pattern is effective and suitable for smoke identification.