Engineering and Applied Sciences Journal

Classification of Images Based on Memorability Score Using Visual Memory Schema and Computer Vision Techniques

Abstract

Harsha Priya Ganapathy and Sankarraj Subramani

Image memorability refers to the intrinsic characteristic of an image that determines how likely it is to be retained in human memory after a brief exposure. Understanding and predicting image memorability is an important problem with applications in computer vision, multimedia retrieval, human-computer interaction, and visual communication. This work presents a computational framework for classifying images based on memorability scores using computer vision techniques and visual memory schema. Large-scale experiments are conducted using natural scene images from the SUN database, where memorability scores are obtained through a controlled human memory game. The study demonstrates strong inter-subject consistency in memorability scores, indicating that memorability is a stable and intrinsic property of images rather than a subjective phenomenon. A detailed analysis of low-level visual features, non-semantic object statistics, and semantic object information is performed to evaluate their influence on memorability. Experimental results show that low-level features alone exhibit weak correlation with memorability, whereas semantic object-level representations significantly improve prediction performance. Support Vector Regression is employed to model memorability scores, and memorability maps are generated to provide interpretable insights into object-level contributions. The proposed approach confirms the critical role of semantic content in human visual memory and supports the development of memory-aware visual systems.

PDF