Total variation denoising method to improve the detection process in IR images
Lina Chato, Shahram Latifi, Pushkin P.E. Kachroo · 2017
Noise reduces the quality of images, resulting in the poor performance of many image processing applications. This paper presents an accurate denoising method to improve the performance of human detection in noisy infrared (IR) images. The local binary pattern (LBP) detector is used to study and analyze the poor effects of noise and to test the efficiency of the proposed denoising method. Noise is removed by minimizing the total variation (TV) of the images, an application that uses partial differential equations (PDEs) and optimal numerical methods to denoise images. The LBP detector shows abnormal behavior with noisy IR images. The Rudin, Osher, and Fatemi (ROF) method based on TV and TV norm1 (TV-L1) methods are efficient in denoising the IR images. In addition, the performance of the LBP detector is improved. Denoising using ROF improves the detection results efficiently as well. Measures for the true positive rate increased from 87% to 89%, and both the false negative per frame and false positive per frame are decreased with respect to the clear IR images. The results prove the efficiency of the proposed method in denoising IR images as well as restoring most of the LBP texture features.