Feature-based active contour model and occluding object detection.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Traditional active contour models (ACM) struggle with occluded objects and noisy images.
• Existing methods often fail in complex scenes with multiple, similarly colored, occluding objects.

Purpose of the Study:

• To develop a robust image segmentation and object detection method.
• To address limitations of existing techniques in handling occlusions and complex scenes.

Main Methods:

• A two-stage approach: 1) Image segmentation using active contour model (ACM) with automatic feature selection (gradient, polarity, depth) and kernel support vector machine (KSVM).
• 2) Salient and occluded object identification using fuzzy C-mean (FCM) algorithm incorporating depth information for robust clustering.

Main Results:

• The proposed method successfully segments and identifies objects, including those in complex occlusions.
• Experimental results on real and standard datasets demonstrate high effectiveness and success.

Conclusions:

• The integrated approach effectively overcomes limitations of standalone ACM and FCM methods.
• Incorporating depth information significantly enhances robustness in segmenting and identifying occluded objects.