High dynamic range infrared image compression and denoising

Detail enhancement and noise removal are major issues in high dynamic range (HDR) infrared image processing. The main focus is how to effectively compress the high dynamic range infrared image to display on the low dynamic range monitors, and in this compression process, both the image details can be enhanced and the noises can be removed. In this paper, we propose a new method to achieve high dynamic range infrared image compression. In the proposed framework, the Laplace differential and Histogram projection are respectively used to sharpen and compress the raw image. Because the image details and noises are enhanced, we design a denoising deep convolutional neural network (DnDCNN)to remove the noises of the image processed in the previous step. By training an additive white Gaussian noise (AWGN) model with different noise levels, our model is capable of processing Gaussian noise at unknown noise levels. Specifically, residual learning and skip connection strategy are utilized to speed up the training process as well as boost the denoising performance. Final extensive experiments demonstrate that our CNN model is very effective in removing such infrared image noise, and the output image has excellent visualization effect.