| 摘要 |
PURPOSE:To color-identify the feature representation of input picture information automatically and at high speed by transferring an n-dimensional feature vector into an m-dimensional finding a histogram and moreover, classifying by the technique of a main component analysis. CONSTITUTION:An n-dimensional feature vector constituting circuit 41 executes statistic processing after converting picture signals R, G and B into the color difference signals of R-Y and B-Y and a Y signal. An n-dimension m- dimension mapping circuit 44 maps in m-dimensional feature space with a main component analysis technique based on a respective axis histogram and covariance matrix on an n-dimensional vector generated by a respective axis histogram generating circuit 42 and a covariance matrix generating circuit 43. Next, respective axis histogram generating circuits 451 to 45m generate the respective dimensional histograms of a m-dimensional feature vector. m Comparators 461 to 46m set threshold values on the m-dimension at every respec tive axis and classify. An encoder 47 puts classified colors together moreover and input picture information can be color-identified automatically and at high speed so that a feature may be represented most effectively. |
