Abstract: An image processing apparatus and a recording medium which can improve the quality of a color document image. Smoothing conditions which are defined by three parameters of “filter size”, “offset”, and “overlap” are set. An edge image which was generated in an edge detection processing is referred to, and a determination is made as to whether an edge pixel exists in a noticed region. When the edge pixel does not exist in the noticed region, a smoothing processing is carried out. When the edge pixel exists in the noticed region, it is determined inappropriate to effect the smoothing processing, and the process goes to a subsequent processing without effecting the smoothing processing. The processing is carried out until the image scanning ends. When the image scanning ends, the next smoothing conditions are set and the processing which is the same as the one described above is performed.

Abstract: An image processing apparatus and a recording medium which can improve the quality of a color document image are provided. An input image is converted into an image having low resolution. A subtractive color image is generated using the converted image having low resolution. From the subtractive color image, adjacent pixels which are allocated to the same representative color are unified so as to extract an FCR (Flat Color Region) candidate region. A region candidate is redetermined using an image having resolution which is higher than the subtractive color image. Next, a boundary of the FCR is detected and an ultimate FCR is determined. Selection of a representative color of the determined FCR is carried out, and a specific color processing which replaces a color which is close to a pure color with the pure color is effected. Finally, image regeneration is carried out by overwriting and drawing (synthesizing) the FCR on the input image.

Abstract: An apparatus and method for recognizing written items including characters and symbols is disclosed. A scanner is utilized for scanning the characters and symbols. A recognition module is used to confidently identify the scanned characters and symbols. The recognition module includes a mechanism for recognizing characters. This mechanism includes a character recognition rule base which yields identified characters and unidentified characters. The unidentified characters are conveyed to neural networks for recognition. The recognition module also includes a mechanism for recognizing symbols. This mechanism includes a symbol recognition rule base which yields identified symbols and unidentified symbols. The unidentified symbols are conveyed to neural networks for recognition. The recognition module also includes a mechanism for context processing of the characters and symbols. The context processor utilizes a multilevel blackboard with a number of ascending levels.

Abstract: A pattern recognition system includes a hierarchical network of parametric pattern recognition components or algorithms of different types. During a "training" phase, distinctions among character types are gathered from a set of correctly labelled training samples. The structure of the component hierarchy is established by recursive training of various subsets of the original training set and, for each component, generation of a "decision function" that either (1) indicates a final classification by the present component the characters of the training, or (2) points to a component lower in the hierarchy, thereby establishing connectivity between components of the hierarchy. The training process continues, generating successively lower components in the hierarchy, until perfect classification is obtained on the training set.