基于熵驱动域适应学习的弱光照图像阴影去除

Shadow Removal of Weak Illumination Image Based on Adaptive Learning in Entropy Driven Domain

针对日常图像获取与处理过程中由于拍摄光线较差等问题,导致图像出现部分阴影,影响图像处理效果等问题,提出基于熵驱动域适应学习的弱光照图像阴影去除方法。通过熵驱动域适应学习技术构建多核分类器,获取弱光照图像的最大平均差异值,完成弱光照图像的预处理。将图像光照分解设定为图像分解与重光照,获取图像最小像素值,根据图像亮度以及RGB方向相似度建立局部像素的约束,完成弱光照图像的分解;采用区域生长法,以检测到的弱光照图像阴影边缘作为基准点,选择灰度值较重部分作为阴影生长起始点,确定弱光照图像的纹理特征值,利用光照补偿方法恢复图像光照,完成弱光照图像阴影去除。仿真结果表明,采用所提方法对弱光照图像阴影去除的效果较好,改善了弱光照图像的质量。

Poor shooting light affects the image processing effect.Therefore, this paper studies the shadow removal method of weak illumination images based on entropy-driven domain adaptive learning.The multi-core classifier was established to obtain the maximum average difference value of weak illumination image via entropy-driven domain adaptive learning technology, thus completing the weak illumination image preprocessing.The decomposition of image illumination was set as image decomposition and re-illumination to get the minimum pixel value of the image.Image brightness and RGB direction similarity were used to establish local pixel constraints in order to complete the decomposition of weak illumination images.Based on the detected edge of shadow in weak illumination image, the region growing method was applied to select the part with heavy gray value as the starting point of shadow growing to determine the texture feature value of weak illumination image.According to the illumination compensation method, the illumination of the image was restored and the shadow of the weak illumination image was removed.The simulation results show that the method has an excellent shadow removal effect and improving image quality.