Selective image segmentation driven by region, edge and saliency functions.

Area of Science:

• Computer Vision
• Image Processing
• Computational Imaging

Background:

• Active contour methods struggle with image segmentation due to texture, color, or intensity variations (inhomogeneities).
• Existing methods face challenges with local minima, slow computation, and weak boundaries, limiting their effectiveness.
• Current techniques often lack the precision needed for complex or real-time image analysis.

Purpose of the Study:

• To develop an advanced image segmentation model that overcomes the limitations of traditional active contour methods.
• To enhance segmentation accuracy for images with inhomogeneous regions and subtle boundaries.
• To introduce a flexible method capable of selective object segmentation.

Main Methods:

• A hybrid approach synchronizing region-based, edge-based, and saliency-based segmentation techniques.
• Utilizing a zero crossing feature detector (ZCD) for edge highlighting and a saliency function for salient region detection.
• Incorporating a globally tuned signed pressure force (SPF) term and level set evolution, with Gaussian kernel for simplified reinitialization.

Main Results:

• The proposed method effectively segments regions with texture, color, and intensity variations.
• Demonstrated capability in precise segmentation of images with weak or subtle boundaries.
• Successfully performs selective object segmentation, allowing for user-defined object selection.

Conclusions:

• The synchronized approach offers a robust and efficient solution for diverse image segmentation challenges.
• The method shows high precision in segmenting natural images, both homogeneous and inhomogeneous.
• Elimination of penalization terms simplifies the level set reinitialization process, enhancing computational efficiency.