一类光照不均图像的特征提取方法

Feature Extraction for a Class of Uneven Illumination Image

图像采集过程中因环境光照不佳等原因往往造成的光照不均,同一物体在不同光照条件下成像差异极大,给图像特征提取带来了挑战。为了提高特征提取对光照不均的鲁棒性,提出了基于色彩衡量的特征检测方法。根据Kubelka-Munk光谱辐射理论,分析计算颜色的空间结构和光谱结构,利用高斯颜色模型估算得到色彩衡量取代灰度图作为信息输入,并在多个尺度下结合Harris角点检测方法进行角点提取,综合得到图像特征信息。实现结果显示,相比传统的特征检测方法,该算法得到的特征点具有数量多、分布均匀及鲁棒性强等优势,较好的解决了光照不均带来的影响。

Changes in lighting are unavoidable and they have a big effect on the way the object looks. It makes a challenge to find a robust local invariant feature descriptor for these uneven illumination images. Most of approaches to feature extraction are based on the premise of that the color image should be converted to grayscale. This paper presents a new approach that it introduces the color independent components based on the Kubelka-Munk model instead of using the gray space to detect the comers. It uses multi-scale Harris comer detection to show the feature of the image, so more details were demonstrated. Experimental results support the potential of the proposed approach.