摘要 |
PROBLEM TO BE SOLVED: To provide a pattern recognizing device robust to the fluctuation of illumination by complementing the weakness of a mutual partial space method to the fluctuation of illumination by a means for localizing/micronizing the fluctuation of illumination, thereby merging both merits with each other. SOLUTION: A pattern input part inputs a reflection luminance pattern of a photographed recognizing object, a fluctuation constancy vector generation part generates a feature vector with which the influence of fluctuation of illumination is localized/miniaturized from each reflection luminance pattern, and an input partial space generation part extracts the characteristic value and an N-dimensional characteristic vector from the N-dimensional feature vectors and generates an input partial space (S1 to S3). An inter-partial space resemblance calculation part calculates the inter-partial space resemblance between the input partial space and its holding dictionary partial space and generates the recognition candidate information with which a pair of category codes of the resemblance and dictionary partial space is defined, a recognition result deciding part compares these resemblance with each other to decide a category code that shows the highest resemblance higher than the prescribed threshold, and an output part outputs the decided category code to the outside (S4 to S6).
|