Entropy- and complexity-constrained classified quantizer design for distributed image classification

In this paper, we address the issue of feature encoding for distributed image classification systems. Such systems often extract a set of features such as color, texture and shape from the raw multimedia data automatically and store them as content descriptors. This content-based metadata supports a wider variety of queries than text-based metadata and thus provides a promising approach for efficient database access and management. When the size of the database becomes large and the number of clients connected to the server increases, the feature data requires a significant amount of storage space and transmission bandwidth. Thus it is useful to devise techniques to compress the features. In this paper, we propose an optimal design of a classified quantizer in a rate-distortion-complexity optimization framework. A Decision Tree Classifier (DTC) is applied to classify the compressed data. We employ the Generalized Breiman, Freidman, Olshen, and Stone (G-BFOS) algorithm to design the optimal pre-classifier, which is a pruned sub-tree of the decision tree, and to perform the optimal bit allocation among classes. The optimization is carried out based not only on a rate budget, but also on a coding complexity constraint. We illustrate this framework by showing a texture classification example. Our results show that by using a classified quantizer to encode the features, we are able to improve the percentage of correct classification by 11%, as compared to using a quantizer without pre-classification at the same rate. This improvement in classification also leads to a reduction of the number of images transmitted between server and client.