An inverse tone mapping operation with two integer data for HDR images

摘要

An inverse tone mapping operation (ITMO) for HDR images with low memory and low computational cost is proposed in this paper. An ITMO generates a high dynamic range (HDR) image from a low dynamic range (LDR) image. It is performed by floating-point operation because HDR images are represented by floating-point data in general. As a result, conventional ITMOs require high computational cost. There are many research works to perform the TMOs with low computational cost. However, most of them are not for inverse TMO, but for forward TMO which converts an HDR image into an LDR image. In order to reduce calculation cost, treat floating-point data as two integer values corresponding to exponent part and mantissa part, and performs ITMO for these integer values separately. In addition, all processes in the method can be performed with fixed-point operation. The experimental results show the proposed method using fixed-point operation is 6.86 times faster compared to the conventional method using floating-point operation.