Visual Recognition Using Local Quantized Patterns

摘要: Features such as Local Binary Patterns (LBP) and Ternary (LTP) have been very successful in a number of areas including texture analysis, face recognition object detection. They are based on the idea that small patterns qualitative local gray-level differences contain great deal information about higher-level image content. Current pattern features use hand-specified codings limited to spatial supports coarse graylevel comparisons. We introduce Quantized (LQP), generalization uses lookup-table-based vector quantization code larger or deeper patterns. LQP inherits some flexibility power visual word representations without sacrificing run-time speed simplicity ones. show it outperforms well-established HOG, LBP LTP their combinations range challenging detection classification problems.