一种从红外序列图像中提取感兴趣区ROI的有效方法

红外序列图像中提取感兴趣区（ROI）是图像压缩的关键步骤，ROI检测质量的好坏直接作用影响着后续处理。文章在Stauffer等人提出的自适应混合高斯背景模型基础上，融入帧差法分别将背景区和运动区以不同的更新率进行更新，同时在自适应背景更新的同时，不断改进自适应学习率。试验表明该方法构建的背景模型有较好的适应性，既可以稳定的对背景进行实时更新，又能够快速的提取目标的感兴趣区。

An Efficient Approach to Extraction ROI from Infrared Image Sequence

ROI（Region of Interest） extraction plays an important role in the field of infrared image signal processing. How to make the compressed image preserve certain auto target recognition performance is an important problem. In order to reduce storage space and lower transmission time, this paper proposed a new algorithm to extract ROI of sequence image. Different from traditional region of interest extraction algorithms, the algorithm extracts region of interest based on a dynamic background modeling approach. Aiming at the slow learning rate of traditional mixture Gaussian model（GMM） this paper proposed a moving object detection algorithm. First, the mixed Gaussian model was constructed, and a new way was adopted to update background which utilized different equations at different phases. With the development of adaptive background update and the adaptive learning rate which always has been improved on the application matures. The validity of the proposed approach is demonstrated on the infrared image ROI extraction. Experimental results show that the approach is efficient both in computational cost and segmentation quality. On the other hand the improved algorithm solves the hole and hollow when the moving object is short of sufficient surface texture by symmetrical differencing algorithm.