Fast Class Rendering Using Multiresolution Classification in Discrete Cosine Transform Domain

To develop a coarse classification scheme which is less dependent on domain-specific knowledge, 2-D discrete cosine transform (DCT) is employed as feature extraction method for vision-based applications. Due to the energy compacting property of DCT, the features of a pattern can be extracted progressively according to their significance. In this paper a multiresolution classification scheme based on DCT is proposed. In coarse classification stage, quantization method is applied to quantize the most significant DCT features, such that the feature space is partitioned into a finite number of grids, each of which corresponds to a grid code (GC). On classifying an unknown object, a reduced set of candidate classes can be retrieved from the corresponding GC. In fine classification stage, the DCT features of the unknown object are extracted progressively according to their importance such that the potential candidate classes which are not well-fitted for the test sample can be pruned as soon as possible. Experiments were conducted for recognizing handwritten characters in Chinese palaeography and showed that our approach performs well in this application domain.