Abstract:

The gray level difference between the pedestrians and environment is small in infrared images, and it is easy to appear the problem of fault segmentation while segmenting. This paper proposes a dual-threshold segmentation method for infrared pedestrian. The global threshold of the image is computed by using the statistical variance, and it is used for a preliminary segmentation. A cross-shaped sliding window is introduced to scan the image. The local threshold of each pixel in the initial segmentation objective area is computed by using statistical variance. By means of the classified formula, the pixel can be classified into objects or the background area. The binary image is obtained. Experiments show that this method improves the segmentation accuracy, and has good performance on pedestrian segmentation.

Key words: infrared image, pedestrian, image segmentation, statistical variance, dual-threshold method, cross-shaped sliding window

摘要：

针对红外图像中行人与环境灰度值相差小、在分割时容易产生错误分割的问题，提出一种双阈值的红外行人分割方法。通过统计方差求解整个图像的全局阈值，对图像进行预分割。设计一种十字形滑动窗口对图像进行扫描，对初始分割得到的目标区域，利用统计方差求解每个像素的局部阈值。使用分类公式将该像素分类为目标或背景区域，得到二值图像。实验结果表明，该方法能提高分割的精确性，对行人的分割效果较好。

关键词: 红外图像, 行人, 图像分割, 统计方差, 双阈值方法, 十字形滑动窗口

CLC Number: 

• TP391.4