Smoke Recognition and Texture Classification Using Improved Local Ternary Patterns

Abstract

To improve detection rate and reduce false alarm rate for smoke recognition, this paper presents local ternary pattern based on confidence level (CLLTP), and further presents a novel multi-CLLTP (M_CLLTP) feature extraction model based on CLLTP. CLLTP is an improved local ternary pattern according to the normal distribution of the pixel values on the difference images. M_CLLTP model computes the CLLTPs for the original images, the weighted CLLTPs for the Gabor feature maps and the CLLTPs for the edge feature maps, and then concatenates the three CLLTP features to generate M_CLLTP feature. Comparative experiments show that M_CLLTP method achieves higher detection rates, lower false alarm rates and higher[F1]scores on three smoke datasets, and the highest mean recall rates on two texture databases. Experimental results indicate that the presented method has a good discriminative ability for smoke and texture, and is very suitable for smoke recognition.

Keywords

• smoke recognition
• local ternary pattern
• confidence level
• gabor transform
• edge feature
• texture classification