联合SSFs选择与图像去马赛克

Joint SSFs selection and image demosaicing

图像去马赛克(Demosaicking)是指将马赛克图像重建为RGB彩色图像的过程,其中马赛克图像的每个像素值只包含一个颜色通道,由单个颜色滤波阵列(Color filter array, CFA)来记录,而去马赛克的过程实际上是对马赛克图像每个像素缺失的两个颜色通道进行估值。目前,研究者们虽然已经证实光谱敏感函数(Spectral sensitivity functions, SSFs)对去马赛克有影响,但是现有的SSFs研究大多只专注于通过深度学习进行联合SSFs设计和去马赛克的相关工作,不能保证所设计的SSFs的物理可制造性。基于此,提出联合SSFs选择与图像去马赛克的深度卷积神经网络,通过使用α-Softmax函数设计卷积层针对图像去马赛克任务对已有的SSFs进行选择,选择出对于去马赛克效果最优的SSFs。研究结果表明:使用α-Softmax函数方法能够通过训练一次选择唯一且最优的SSFs,联合方法对去马赛克效果有促进作用。因此,在设计CFA时可以通过使用α-Softmax函数首先选择出最优SSFs,以增强重建图像质量、降低制作成本。

Image demosaicking(Demosaicking)refers to the process of reconstructing a mosaic image into an RGB color image,in which each pixel value of the mosaic image contains only one color channel,which is recorded by a single color filter array(Color filter array,CFA).In the demosaicing process,the missing two color channels of each pixel of the mosaic image will be evaluated actually.At present,researchers have confirmed that Spectral sensitivity functions(Spectral sensitivity functions,SSFs)have an impact on demosaicing,but most of the existing research on SSFs only focuses on joint SSFs design and demosaicing through deep learning,the physical manufacturability of the designed SSFs cannot be guaranteed.Therefore,based on the current situation,a deep convolutional neural network that combines SSFs selection and image demosaicing is proposed.By using theα-Softmax function to design the convolutional layer,the existing SSFs are selected for the image demosaicing task,and the best demosaicing effect is selected.The research shows that using theα-Softmax function method can select the unique and optimal SSFs through training once,the joint method can promote the demosaicing effect.Therefore,when designing CFA,the optimal SSFs can be selected first by using theα-Softmax function in order to enhance the reconstructed image quality and reduce production costs.