Comparative Sharpness Evaluation for Mobile Phone Photos.

Qiang Lu,Guangtao Zhai,Yucheng Zhu,Xiongkuo Min,Tao Wang,Xiao-Ping Zhang

DOI: https://doi.org/10.1007/978-3-030-93046-2_1

2021-01-01

Abstract:Mobile phones are the main source of a vast majority of digital photos nowadays. Photos taken by current mobile phones generally have fairly good visual quality without noticeable distortion. This progress benefits not only from the higher specification of camera sensor, but also the excellent imaging algorithm. Reliable and practical photo quality evaluation algorithm can guide the development of mobile phone photography to a higher level, which is also the key to this progress. Clearly, subjective photo quality assessment suffers from the drawbacks of productivity and reproducibility. Traditional objective image quality assessment algorithms can hardly discern the subtle quality difference between pictures photoed by different mobile phones. In this paper, we propose a comparative sharpness evaluation method based on an improved Siamese network with multilayer features for mobile phone photos. We employ Resnet-18 as the main feature extraction module. The features extracted from different layers of the two branches will be concatenated layer by layer and then passed to the final fully connected layer. The final output represents the sharpness comparison result of the two input pictures. Experimental results show that our sharpness evaluation algorithm achieves state-of-the-art performance on the SCPQD2020 database.