Linear Predictive Model for Discriminative Feature Representation of Object Classification
Ekachai Phaisangittisagul · 2018
In order to improve the performance for recognition in computer vision tasks, a high-level feature representation plays a crucial role to transform a raw input data (low-level) into a new informative representation for learning algorithms. Sparse coding is one of the widely used methods to generate a high-level feature representation for classification. In particular, an input data can be represented as a sparse linear combination of a set of training overcomplete dictionary. However, the main problem in traditional sparse coding is that it is fairly slow to compute the corresponding coding coefficients due to an ℓ0/ℓ1optimization. In this work, an efficient linear model with low computational effort is proposed to create the discriminative coding coefficients. The comparison of classification performance between the proposed method and the existing discriminative sparse coding is experimented on image databases for face and scene recognitions under the same learning condition. The results indicate that our proposed method both achieves promising classification accuracies and outperforms in computation time.