Blind Tone-Mapped Image Quality Assessment Based on Brightest/Darkest Regions, Naturalness and Aesthetics

Abstract

A tone-mapped image (TMI) converted from its high-dynamic-range image (HDRI) tends to appear overexposed in its brightest regions or underexposed in its darkest regions, resulting in inevitable loss of details and impaired naturalness and aesthetics. To address this issue, this paper proposes a novel blind TMI quality assessment (BTMIQA) method for HDRIs used in different dynamic range displays. The brightest and darkest regions of the TMI are first defined and segmented, and their local detail features are used in combination with the global detail feature of the TMI to evaluate the detail distortion. The natural scene statistics features of the luminance and yellow channels of the TMI are then used to evaluate the luminance naturalness and chrominance naturalness, respectively. Subsequently, to predict the aesthetics of the TMI, a series of colorfulness features are extracted and concatenated in a final feature vector, which is used to evaluate the TMI quality by random forest regression. Experiments are performed on the TMI database and ESPL-LIVE high-dynamic-range database, and the results show the effectiveness of the proposed method. Compared with the existing TMI quality assessment methods, the proposed BTMIQA method is more consistent with human visual perception.