Codes for the paper submitted to ICIP2016 titled "completed local derivative pattern for rotation invariant texture classification" By Yuting (huyuting@gatech.edu) Label: cldp_version1.0 Date: 02/02/2016.
Publication: Y. Hu, Z. Long, G. AlRegib, “Completed Local Derivative Pattern for Rotation Invariant Texture Classification,” IEEE International Conference on Image Processing (ICIP 2016), pp. 3548-3552, 2016.
Bib: @inproceedings{hu2016completed, title={Completed local derivative pattern for rotation invariant texture classification}, author={Hu, Yuting and Long, Zhiling and AlRegib, Ghassan}, booktitle={Image Processing (ICIP), 2016 IEEE International Conference on}, pages={3548--3552}, year={2016}, organization={IEEE} }
-
Resources
The codes are based on the following links:
- CLBP: the source codes of the proposed algorithm can be downloaded from http://www.comp.polyu.edu.hk/~cslzhang/code/CLBP.rar. The implemnetation of binary pattern extraction (CLDP) part is based on the revision of CLBP codes.
- DRLBP(Dominant Rotated Local binary pattern) MatlabCode: http://www.cs.tut.fi/~mehta/drlbp. The implementation of the CLDP-based texture classification on the Outex database is based on the classification part in DRLBP codes.
-
Objective, database and main functions
- Objective: texture classification based on the histogram fusion of completed local derivative patterns (CLDP)
- Database: three test suits => Outex TC10("inca"), TC12("t184" and "horizon")
- Main functions a. cldp_d(img,r,p,mapping,'h'): inputs: original image img, radius r, the number of sampoing points p, mapping manners mapping, hitogram 'h' outputs: the cldp_d histogram of an image b. test_cldp_d_outex=test_clbp_d_outex(p,r): inputs: radius r, the number of sampoing points p outputs: classification accuray (CA) based on nearest neighbor classifier (NNC), compuataion time under a certain (P, R) for one test suit c. test_cldp: outputs: test cldp-based classification performance on each test suit using various combinations of (P, R) values d. getmapping: three selections: 'u2' for uniform 2, 'ri' for rotation-invariant, 'riu2' for uniform 2 and rotatoin-invariant e. ClassifyOnNN(DM,trainClassIDs,testClassIDs): input: "DM" is a m*n distance matrix, m is the number of test samples, n is the number of training samples, 'trainClassIDs' and 'testClassIDs' stores the class ID of training and test samples output: classifcation accuracy
-
Results reproduction PS: we run Matlab2014b codes on Intel Core i7-4790K 4.00GHz
- Original experimental results: ...\CLDP\Outex\Results\CDLP-based Classification Results_20160128.xlsx a. 1st subsheet (CLBP VS CLDP): comparisons between CLBP and CLDP on classification accuracy for each combination scheme under a certain (P, R) Each row represents a combination scheme and each colomn represents a certain (P, R) for a test suit. For example: CLDP_S/M/D/C(riu2): run matlab codes under ...\CLDP\Outex\Test_TC12\Test_TC12_000\Test_cldp\Test_cldp_smdcjoint; The codes return accuracy for the CLDP_S/M/D/C-based texture classifictition under various (P, R) values on test suit TC12_000("t184"). b. 2nd subsheet (CLBP VS CLDP): comparisons between CLBP and CLDP on computation time for each combination scheme under a certain (P, R)
- Table 1: Average classification accuracy (%) of CLBP and CLDP on TC10 ("inca") and TC12 ("t184" and "horizon") using different combination schemes This table can be found from the 1st subsheet in ...\CLDP\Outex\Results\CDLP-based Classification Results_20160128.xlsx
- Table 2: Classification accuracy (%) of state-of-the-art texture descriptors on TC10 ("inca") and TC12 ("t184" and "horizon") This table can be found from the 4th subsheet in ...\CLDP\Outex\Results\CDLP-based Classification Results_20160128.xlsx