Detecting Salient Image Objects Using Color Histogram Clustering for Region Granularity

Area of Science:

• Computer Vision
• Image Processing

Background:

• Salient object detection is crucial for computer vision applications but challenging due to complex color image properties.
• Existing region-based methods struggle with optimal region granularity, leading to segmentation issues.

Purpose of the Study:

• To develop a robust and efficient method for salient object detection in color images.
• To overcome the limitations of existing region-based approaches by automating region granularity selection.

Main Methods:

• Utilized color histogram clustering to automatically identify homogenous regions.
• Computed region saliency based on color contrast, contrast ratio, spatial features, and center prior.
• Applied morphological operations to refine saliency maps and remove artifacts.

Main Results:

• The proposed method effectively detects salient objects by integrating multiple saliency cues.
• Experimental results demonstrate superior performance compared to 30 non-deep learning and 7 deep learning methods.
• The approach is computationally efficient and robust across diverse image categories.

Conclusions:

• Introduced a novel, simple, and efficient color histogram clustering method for salient object detection.
• The method addresses the critical issue of region granularity selection in saliency computation.
• The proposed technique offers a significant advancement in accurate and efficient salient object detection.