Predicting Region of Interest in Human Visual Search Based on Statistical Texture and Gabor Features

TL;DR

This work addresses predicting human fixations during location-unknown visual search in medical images by fusing texture statistics and local orientation information. It introduces two pipelines that combine gray-level co-occurrence matrix (GLCM) texture features with 2D Gabor features within a Gaussian mixture model framework and validates them on simulated digital breast tomosynthesis (DBT) data generated with VICTRE, with an additional thresholded-data baseline. A key finding is a strong positive correlation between GLCM mean and Gabor responses, $r=0.765$, indicating these features convey related image structure, and eye-tracking data show predicted regions align with early gaze. The results imply that integrating texture-based and structure-based cues yields more accurate, perceptually grounded observer models for location-unknown search tasks in medical imaging.

Abstract

Understanding human visual search behavior is a fundamental problem in vision science and computer vision, with direct implications for modeling how observers allocate attention in location-unknown search tasks. In this study, we investigate the relationship between Gabor-based features and gray-level co-occurrence matrix (GLCM) based texture features in modeling early-stage visual search behavior. Two feature-combination pipelines are proposed to integrate Gabor and GLCM features for narrowing the region of possible human fixations. The pipelines are evaluated using simulated digital breast tomosynthesis images. Results show qualitative agreement among fixation candidates predicted by the proposed pipelines and a threshold-based model observer. A strong correlation is observed between GLCM mean and Gabor feature responses, indicating that these features encode related image information despite their different formulations. Eye-tracking data from human observers further suggest consistency between predicted fixation regions and early-stage gaze behavior. These findings highlight the value of combining structural and texture-based features for modeling visual search and support the development of perceptually informed observer models.

Figures (11)

• Figure 1: Examples of 2D DBT slices of breast phantoms with different density categories.
• Figure 2: Segmentation based on texture features.
• Figure 3: Mask based on texture features.
• Figure 4: Clustered candidates based on texture features.
• Figure 5: Initial candidates from pipeline B.