摘要 |
PURPOSE:To decrease the capacity of a dictionary by grouping components of a feature vector for each fixed dimension number, quantizing the components in each group with the quantizing level number of each dimension to obtain quantization data of a fixed number of bits for each group, and obtaining a compressed vector containing said quantization data as a component. CONSTITUTION:A rearrangement table 7 obtains the sum of the mean value of each dimension of a feature vector and standard deviation when a dictionary 11 is produced and is produced based on the width of said sum. A rearrangement part 8 refers to the table 7 to rearrange the dimensions of the feature vectors inputted from a feature extracting part 5 and inputs successively 4-dimensional groups to a quantizing part 10. The part 10 uses the component value (histogram value) of each dimension of each input group to access a quantizing level table 9 to read out the quantization data of a single byte. Thus the feature vector and the dictionary are effectively compressed so that the dictionary capacity is extremely decreased.
|