Research On Illumination Consistency Based On Single Low Dynamic Range Image

Augmented Reality occupies an important position in the field of computer graphics,it will overlay computer-generated virtual scene to the real scene to make them look like awhole, which make people interact with virtual objects in the scene more naturally. Tomake scenes synthesized with augmented reality seem realistic, illumination consistencymust be considered.A general framework of illumination consistency based on single low dynamic rangeimage is present. Firstly, the low dynamic range image must be preprocessed with waveletdenoising. Then we expand the dynamic range of image to generate high dynamic rangeimage. Finally, relighting virtual objects with high dynamic range image. Waveletdenoising based on Gaussian mixture model is introduced for image in the pro-processingmodule, which can not only remove noise effectively, but also avoid the distortion of theedges and other details of image. In the module for the generation of high dynamic rangeimage, image is linearized with inverse response function of camera, the luminance rangeof pixel is expanded according to inverse tone mapping function. In order to compensatefor the losing highlight information of the image, expand map is calculated with a methodcombinded with thresholding and flooding Gaussian smoothing. Linear interpolation isperformed with expand map to obtain a high dynamic range image. Then, relightingvirtual object to achieve illumination consistency.Detial experiment and analysis was performed for the framework. In order to verifythe performance of denoising algorithm and the accuracy of the inverse tone mappingfunction of camera, several other methods was adopted respectly to compare and analy theresult. The method based on single low dynamic single image can greatly reduce the timecost of image exposure and processing and lower the limit of photographic equipment,impoving the practicality and real-time of the algorithm at the same time.