Face Recognition Using Multiresolution Hybrid Kekre-DCT Wavelet Transform Features with Multiclass ECOC Framework
Alpa Choudhary, Rekha Vig · Procedia Computer Science · 2018
Error Correcting Output Codes (ECOC) framework is efficient approach for multiclass classification problems by reducing them to binary classifiers. In this paper Face recognition system is implemented by extracting features using novel multiresolution hybrid wavelet transform approach. This approach combines two orthogonal transform like Kekre and DCT transform using Kronceker product to generate hybrid wavelet transform which gives different levels of resolution. The generated transform also uses energy compaction technique to obtain dimensionality reduction. The extracted features are then passed for classification via ECOC approach where binary classifiers are coded using Sparse random coding design and decoded using Hamming distance. The final model predicts the class labels. The best accuracy of 99.67% is obtained at 99% of energy threshold.