Research On Tone Mapping For High Dynamic Range Images

High dynamic range images which have rich details, are more realistic than general images and widely used in many fields.In order to solve the problem of the mismatch of HDR images and display devices' dynamic-range, researchers proposed many tone mapping algorithms for high dynamic range images, but there are still some problems to be solve. This thesis systematically studies tone mapping algorithms for high dynamic range images. In order to solve the problem of halo artifacts and noise enhanced in the processed image, this thesis proposes two new tone mapping algorithms. The main work and achievements of this thesis can be summarized as follows:1. Studied, analyzed and compared the existing edge-preserving smoothing filters:fast bilateral filter, weighted least square filter, guided filter and L~0 smoothing filter, and analyzed the causes of halo artifacts theoretically. In the layer-decomposition-based tone mapping algorithms, the use of Gauss filter will introduce halo artifacts along image highcontrast edges, the problem can be solve by using edge-preserving smoothing filter.2. Proposed a novel local tone mapping algorithm base on L~0 smoothing filter, which can effectively reduce halo artifacts while with good contrast and detail preservation. The method imitates the adaptation mechanism of the human visual system, consists of global tone mapping and local tone mapping. In the local tone mapping, the HDR image is decomposed into two layers through L~0 smoothing filter: base layer and detail layer, then the base layer which reflects the change of environment illumination is compressed.3. Proposed a layer-decomposition-based contrast limited adaptive histogram equalization algorithm, which can effectively solve the problem of noise enhanced. The contrast limited adaptive histogram equalization can compress the dynamic-range of image,but it will enhance the noise of HDR image. This thesis argues that the high frequency component contain the details and noise of image, and the low frequency component reflects the change of illumination, therefore, high frequency component and low frequency component is processed respectively, and it achieves good results.