A unified tone mapping operation for HDR images including both floating-point and integer data

Abstract

This paper considers a unified tone mapping operation (TMO) for HDR images. This paper includes not only floating-point data but also long-integer (i.e. longer than 8-bit) data as HDR image expression. A TMO generates a low dynamic range (LDR) image from a high dynamic range (HDR) image by compressing its dynamic range. A unified TMO can perform tone mapping for various HDR image formats with a single common TMO. The integer TMO which can perform unified tone mapping by converting an input HDR image into an intermediate format was proposed. This method can be executed efficiently with low memory and low performance processor. However, only floatingpoint HDR image formats have been considered in the unified TMO. In other words, a long-integer which is one of the HDR image formats has not been considered in the unified TMO. This paper extends the unified TMO to a long-integer format. Thereby, the unified TMO for all possible HDR image formats can be realized. The proposed method ventures to convert a long-integer number into a floating-point number, and treats it as two 8-bit integer numbers which correspond to its exponent part and mantissa part. These two integer numbers are applied the tone mapping separately. The experimental results shows the proposed method is effective for an integer format in terms of the resources such as the computational cost and the memory cost.