| Recently, high dynamic range images have caused more and more attention from people. And it is becoming more and more important and common in the digital image processing field. With the rapid development of the hardware technology, the high dynamic range images of real scenes are much easier to capture than before. However,most of display devices can only support low dynamic range output, which can not reproduce the high dynamic range image realistic.In order to solve the problem, researchers proposed many tone mapping algorithms for high dynamic range images. These algorithms are based on the spatial domain,which are divided into spatially uniform algorithms(global algorithms) and spatially varying algorithms(local algorithms). The global algorithm is simple, which can keep the fine effect of darker and brighter area. But it may cause detail loss during the processing. And it could not handle complex scenes. The local algorithm can better simulate human vision system and preserve detail information and image contrast. But it is computational expensive and the halo effect will appear around edges after tone mapping.This paper introduces the principle and related knowledge of HDR tone mapping algorithms. Some typical methods are also discussed. We introduce the multi-scale image decomposition based on filters and iCAM06 model and its application in tone mapping.To solve the halo effect around edges after tone mapping, a multi-scale tone mapping approach using non-local means filter is presented. The proposed approach produces a series of coarsening images. The original image is decomposed into a base layer, containing larger scale variations in intensity, and some detail layers capturing the smaller scale details. The base layer and detail layer are dealt with different strategies respectively and followed by a restoration of color information. his algorithm can not only avoid halo artifacts, but also preserve more details efficiently while reproducing color information in visual point of view.In order to reduce computational expenses and reconstruct color information and tone mapping. A regionalized non-local means filter is proposed. Together with imagecolor appearance model(i CAM) framework. The original non-local means filter is modified by considering region segmentation as a new factor for the weight function.The new filter is employed to decompose image into base layer and detail layer. The reconstructed image is obtained by several steps including chromatic adaptation, tone compression, detail compensation and adjustment. Experimental results show that the model can preserve the details, accurately predict and restore the color information. |
