基于深度学习的视觉图像非显著性区域增强

Non significant area enhancement of visual image based on deep learning

视觉图像的非显著性区域内存在多种低照度区域,易导致图像输出灰度级偏大、丢失的图像细节较多。为此,该研究设计了基于深度学习的视觉图像非显著性区域增强方法。首先采用多尺度高斯函数在图像区域内设置自适应参数,控制明暗不一的图像区域产生的减弱效应,从而提取非显著区域光照分量。然后利用深度学习过程构建非显著区域中的数值成像模型并重建图像分辨率,再通过定义引导函数构建非显著区域分层增强算法。实验结果表明:相比于传统方法,应用文中方法后,图像输出灰度值更小、丢失的图像细节更少。

There are many low light areas in the non significant area of visual image, which is easy to lead to large gray level of image output and more lost image details. Therefore, this study designs a non significant region enhancement method of visual image based on deep learning. Firstly, multi-scale Gaussian function is used to set adaptive parameters in the image area to control the attenuation effect in the image area with different light and shade, so as to extract the light component in the non significant area. Then, the deep learning process is used to construct the numerical imaging model in the non significant area and reconstruct the image resolution, and then the hierarchical enhancement algorithm of the non significant area is constructed by defining the guiding function. The experiment results show that compared with the traditional methods, the output gray value of the image is smaller and the lost image details are less.