Meaningful matches in stereovision

Area of Science:

• Computer Vision
• Image Processing
• Statistical Analysis

Background:

• Reliable block matching is crucial for image analysis tasks.
• Existing methods may struggle with chance occurrences and periodic patterns.
• The need for robust statistical validation in image block comparison.

Purpose of the Study:

• To introduce a novel statistical method for reliable image block matching.
• To ensure that identified block matches are statistically significant and not coincidental.
• To provide a quantifiable measure of match reliability based on false alarm rates.

Main Methods:

• Development of a statistical background model learned from image data.
• Application of an a contrario approach to guarantee a controlled number of false alarms.
• Integration of a parameterless self-similarity threshold to handle periodic objects.
• Experimental validation on datasets with occlusions and non-simultaneous stereo imagery.

Main Results:

• The proposed method reliably distinguishes true matches from chance occurrences.
• A fixed number of false alarms (false positives) is guaranteed on average per image.
• Match reliability is directly measured by the associated number of false alarms.
• The method effectively detects occlusions and incoherent motions in non-simultaneous stereo images.
• The self-similarity threshold successfully complements the a contrario method for periodic patterns.

Conclusions:

• The introduced statistical block-matching method offers a reliable and statistically validated approach to image comparison.
• The method's parameter (false alarm number) provides a direct measure of match reliability.
• The combination with a self-similarity threshold enhances robustness against periodic objects.
• The technique demonstrates practical utility in detecting complex scene dynamics like occlusions and motion in stereo vision.