| The matter of generating high dynamic range (HDR) imaging arises to satisfy the needs of industrial applications and high-quality imaging. A number of HDR image generation algorithms have been proposed in the past, including the hardware methods and software methods. However, considering the increasse of cost in haftware and the versability in applications, we focus on software methods in this project, that is, compose the high dynamic range radiance map from the input image sequence under different exposures.However, traditional software high dynamic range image generation algorithms cannot completely exclude the negative influences of the noisy pixels in the input images on the output image, and are thus not able to produce results with highest possible SNR, which is especially critical in industrial applications. In this paper we introduce a new HDR generation algorithm based on the multi-exposure Retinex model deduced in this paper for HDR image composition. Compared with the traditional algorithms, the result image produced by the new algorithm is of better quality, with higher SNR and a dynamic range no less than those produced by traditional methods.Due the different properties of the luminance images and reflectance images, L and R are synthesized indepently before being combined together. For the composition of the luminance component, because it is of high dynamic range and has low noise levels, traditional HDR image composition method can be appied. But for the reflectance component, because it is includes the intrinsic properties of the objects, R is synthesized by a novel method. All the most reliable pixels with highest SNR are collected in the output to produce the highest possible SNR in the HDR image generated.The newly proposed HDR image generation algorithm can not only be applied to the grey-level images typical in industrial applications, but also on color images of normal formats and raw formats based on the methods mentioned in this paper. |
PDF Full Text Request