Adaptive color segmentation-a comparison of neural and statistical methods

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Advancements in computing power enable pixel-based operations on real-time color images.
• Adaptive classification tools, such as neural networks, facilitate the development of specialized object detectors.
• These detectors can segment complex objects after training on labeled images.

Purpose of the Study:

• To compare a neural network approach using local linear maps (LLMs) against a normal distribution classifier.
• To evaluate an adaptive segmentation method utilizing local color information for object and background classification.
• To apply the method for recognizing and localizing human hands in complex laboratory scenes.

Main Methods:

• Implementation of a neural network model based on local linear maps (LLMs).
• Development of a comparative classifier utilizing normal distributions.
• Utilizing local color information to estimate class membership probabilities for adaptive segmentation.

Main Results:

• The neural approach based on LLMs was detailed and compared with a normal distribution classifier.
• The adaptive segmentation method effectively uses local color information.
• Successful application to human hand recognition and localization in complex scenes.

Conclusions:

• The study provides a detailed comparison of two adaptive segmentation techniques.
• The proposed method demonstrates effectiveness in object recognition tasks within complex visual environments.
• Neural networks, particularly LLMs, offer a viable approach for sophisticated image segmentation and object detection.