Improved Dictionary Learning Scheme with Cross-Label for Scene Classification

In recent decades, dictionary learning has been attracted strong attentions due to its great performances when applied in many applications such as signal reconstruction and scene classification. Conventional dictionary learning schemes have been developed to the scene classification but it hard to make a suitable tradeoff between accuracy and efficiency. This paper proposed an improved method of dictionary learning with crosslabel and group regularization, in order to achieve a tradeoff between accuracy of classification and efficiency of algorithm execution. Demonstrated by the experiment results on the Scene15 dataset. Our proposed approach method is obviously improved owing to the algorithm can not only obtain a desired classification performance, but also decrease much of the computational time during the process of dictionary learning.