Matlab implementation, comparision and improvement of Local texture descriptors. This repo demonstrate usage of Local binary pattern (LBP), Local derivative pattern (LDP), Local Tetra pattern (LTrP), Noise Resistant LBP (NR-LBP), Histogram Refinement of Local texture descriptor for Content based image retrieval (CBIR) application.
This repo has matlab implementation of paper titled "Histogram refinement for texture descriptor based image retrieval"
Avalable at https://www.sciencedirect.com/science/article/pii/S0923596517300164. Cite the above paper if using code/files from this repo.
- Download COREL 1000 dataset from and extract all images inside "image.orig" folder.
- Run cbir.m
- Run Run_benchmark.m
- Use the results generated for benchmarking.
The result from Run_benchmark.m is shown below which is very close to the values reported in the paper.
Generating Local Skew pattern (LSP); Percentage completed : 1000 / 1000
benchmarking Local biniary pattern (LBP); Percentage completed : 1000 / 1000
classification accuracy: LBP (original) is 69.120
classification accuracy: LBP (With histogram refinement) is 74.910
benchmarking Local derivative pattern (LDP); Percentage completed : 1000 / 1000
classification accuracy: LDP (original) is 69.600
classification accuracy: LDP (With histogram refinement) is 74.380
benchmarking Local Tetra pattern (LTrP); Percentage completed : 1000 / 1000
classification accuracy: LTrP (original) is 65.820
classification accuracy: LTrP (With histogram refinement) is 71.000
benchmarking Noise-resistant Local biniary pattern (NRLBP); Percentage completed : 1000 / 1000
classification accuracy: NRLBP (original) is 67.450
classification accuracy: NRLBP (With histogram refinement) is 73.540
The sample results presented below are generated from COREL dataset where the first column is the query/search image and the rest of the column are top matches found (in order) based on L1 similarity measure of local texture histogram.
