Robust color histogram descriptors for video segment retrieval and identification.

Area of Science:

• Computer Vision
• Image Processing
• Multimedia Systems

Background:

• Representing color features across multiple video frames (GoF) is crucial for visual information management but challenging.
• Key frame-based methods often yield unreliable results due to dependency on representative frame selection.
• Existing methods struggle with variations like brightness changes, occlusion, and editing effects.

Purpose of the Study:

• To develop reliable and efficient histogram-based color descriptors for representing color properties of multiple images or GoF.
• To introduce alpha-trimmed average histograms and intersection histograms for improved color feature representation.
• To enhance video segment retrieval and query frame identification.

Main Methods:

• Developed alpha-trimmed average histograms by filtering individual frame histograms to create robust color representations.
• Introduced the intersection histogram to identify common pixel colors across all frames in a GoF.
• Applied these descriptors to video segment retrieval and query frame identification algorithms.

Main Results:

• Alpha-trimmed average histograms demonstrated superior performance over key frame-based methods in video segment retrieval.
• The intersection histogram enabled a fast and efficient algorithm for query frame belonging identification.
• Proposed descriptors were validated and included in the ISO standard MPEG-7.

Conclusions:

• Histogram-based color descriptors offer a robust and efficient solution for representing color features in visual information management.
• The developed methods significantly improve the accuracy and reliability of video analysis tasks.
• Inclusion in the MPEG-7 standard highlights the practical significance and effectiveness of these descriptors.