Research On Infrared Image Enhancement Technology Based On Sparse Representation And Retinex Theory

Infrared image system is based on the infrared radiation difference between the target and the background in the image, which is widely used in both military and industry fields for its advantage of high temperature sensitivity, high dynamic range, high capacity to work in fog and capacity of working around the clock. Compared with the CCD image, infrared image has lower contrast, lower SNR and is always dimmer, which is caused by the degradation during the imaging process and the transmission process. These features will affect the visual ability and following operate, such as the image interpretation and target extraction on eyes. So we should first solve the image enhancement problems such as image detail enhancement and super-resolution problem.We first studied the imaging mechanism and the infrared image specification, and analysed the properties of the infrared images, which will be used in the later study. We also studied the image quality assessment which is consist of the subjective assessment and the objective assessment. And we studied the objective indicators, such as contrast, information entropy, standard deviation, which is used to assess the image quality after the image enhancement and super-resolution.In the paper, we studied the conventional method of the infrared image detail enhancement and analysed the visual effect of the methods. Then we proposed a method based on Retinex theory and adaptive gain control which can enlarge the visual dynamic range of the image and make the details in the small regions more clearly. In our method, we first depart the image into high frequency part and low frequency part by filtering; and process the low frequency part with the Multi-Scale Retinex method to enlarge the visual dynamic range of the image; and process CLAHE method on the high frequency method to make the image details more clearly; finally, we add them up with their weight value and get the final output image. Experiment results shows that our algorithm can successively enhance the details of the image and the output images have high contrast, high dynamic range and the visual effect is improved significantly.We also studied the conventional method of the infrared image super-resolution. Contraposing the main problems of the conventional method, such as the computational complexity and the lack use of the prior information, we proposed a new super-resolution algorithm combining with the image sparse representation theory. In our algorithm, we bring the image sparse representation restrain into the degradation model to reconstruct the high resolution image. We use the OMP algorithm to seek the optimal solution of each image block; then we put the reconstructed high resolution blocks together and calculate the average value on the neighborhood pixels. Experiment shows that our method can reconstruct the image details and improve the visual effects and our method has better effect compared with the interpolation methods.