Individual differences in image preferences: a personalized image enhancement method

Poster Presentation 43.348: Monday, May 22, 2023, 8:30 am – 12:30 pm, Banyan Breezeway
Session: Image Preference, Statistics and Aesthetics

Sarit F. A. Szpiro1,2 (), Amit Yashar1,2; 1Special Education Department, University of Haifa, 2The Edmond J. Safra Brain Research Center, University of Haifa

INTRODUCTION: A person's light sensitivity varies depending on lighting conditions and the state of their sensory system. Yet, current image enhancement algorithms do not consider individual differences, instead offering a "one size fits all". For example, histogram equalization improves the visibility of nighttime images by modifying the luminance histogram to the full scale between black and white. Here, we developed a new algorithm that enhances images according to individual preferences and demonstrate its efficacy in an experiment. METHODS: Phase 1 - image adjustment: for each image (5 in daylight and 13 in nighttime), participants (n=12) moved the mouse in the x and y directions to simultaneously adjust two parameters, the mean and variance of a gaussian function that determined the luminance and contrast of the image. Images were adjusted twice-when presented alone or presented near the histogram equalized image. Phase 2 – discrimination: we tested whether the selected parameters in phase 1 improved the visibility of the adjusted image. Each adjusted image was compared to the original image or to the histogram equalized image, and the participant selected the image that appeared to have more details. RESULTS: Phase 1: Image parameters varied across image type and participants. Importantly there was a high correlation between the first and second image adjustment both across images and participants. These finding suggests individual difference in image enhancement preferences. Phase 2: With daylight images, participants choices did not differ from chance, that is they chose adjusted or comparison images randomly. However, when nighttime images were presented, participants preferred the adjusted images significantly more than the original and the histogram equalized images (88% and 74%, respectively). Our study quantifies variation in image enhancement preferences across neurotypical individuals, demonstrates the efficacy of our method, and that individual differences should be considered in future applications of image processing.