LBP-based edge-texture features for object recognition.

Area of Science:

• Computer Vision
• Image Processing
• Machine Learning

Background:

• Existing Local Binary Pattern (LBP) and Local Ternary Pattern (LTP) methods struggle with object-background contrast discrimination.
• Robust LBP (RLBP) has limitations in mapping LBP codes and their complements.
• Current texture features often discard essential contrast information for object contour representation.

Purpose of the Study:

• To propose novel edge-texture features, Discriminative Robust Local Binary Pattern (DRLBP) and Ternary Pattern (DRLTP).
• To address limitations in discrimination and contrast information retention of existing LBP, LTP, and RLBP.
• To enhance object recognition accuracy through improved feature representation.

Main Methods:

• Development of Discriminative Robust Local Binary Pattern (DRLBP) and Ternary Pattern (DRLTP) features.
• Comparative analysis against LBP, LTP, and RLBP on challenging datasets.
• Evaluation of feature performance in object recognition tasks.

Main Results:

• DRLBP and DRLTP effectively solve discrimination issues between bright objects on dark backgrounds and vice-versa.
• DRLBP resolves the complement mapping problem present in RLBP.
• The proposed features successfully retain contrast information, crucial for object contour representation.
• Outperformance of DRLBP and DRLTP over existing methods on multiple benchmark datasets.

Conclusions:

• DRLBP and DRLTP represent significant advancements in edge-texture feature extraction for object recognition.
• The proposed features offer superior discrimination and contour representation capabilities.
• These novel features provide a more robust and accurate approach to object recognition tasks.