Abstract
In recent years, the remarkable progress of imaging systems has made it possible to capture the high dynamic range images, while the current display devices still have the low dynamic range. This paper proposes a novel method for dynamic range compression in captured images with keeping the input/output visual contrast by taking the ratio of a pixel to its local surround average based on Center/Surround Retinex model. The proposed method has the following features: [1] A novel spatially-variant LCRT (Local Contrast Range Transform) operator preserves local visual contrast of output image as much as that of input image, [2] The gradation and luminance in range transformed regions are reproduced smoothly and naturally after compression, and [3] LCRT is designed using a SS (Single-Scale) Gaussian surround but works in very stable without banding artifacts even near the boundaries of heavy changes in shadow and highlight. The paper discusses how the LCRT works stable different from Retinex though using the same SS surround from a theoretical point of view and demonstrates the features and the performances based on the comparative experiments with Center/Surround Retinex model.
