DOCUMENT RECOGNITION DEVICE AND DOCUMENT RECOGNITION METHOD

To provide a recognition engine that dynamically selects an optimum discriminator on the basis of an operation condition.SOLUTION: A document recognition device comprises: a control section that selects a character string recognition section for recognizing a character string included in an input image from a plurality of character string recognition sections for recognizing a character string in an image; an operation state storage section that stores values of a plurality of evaluation indices obtained by evaluating results of character string recognition executed in the past; and an operation assumption value storage section that stores the values of the plurality of evaluation indices assumed in each character string recognition section. When an operation condition including the values of the plurality of evaluation indices required for the character string recognition to be executed is input, the control section selects any one of the plurality of character string recognition sections so that the values of the plurality of evaluation indices for evaluating the results of character string recognition to be executed approximate a value satisfying the operation condition on the basis of the operation condition, the values of the plurality of evaluation indices stored in the operation state storage section, and the values of the plurality of evaluation indices stored in the operation assumption value storage section.SELECTED DRAWING: Figure 1 【課題】運用条件にもとづいて最適な識別器が動的に選択される認識エンジンを提供する。【解決手段】画像中の文字列を認識する複数の文字列認識部から、入力された画像に含まれる文字列の認識を実行する文字列認識部を選択する制御部と、過去に実行された文字列の認識の結果を評価した複数の評価指標の値を格納する運用状態格納部と、各文字列認識部において想定される複数の評価指標の値を格納する運用想定値格納部と、を有する文書認識装置であって、制御部は、実行される文字列の認識に要求される複数の評価指標の値を含む運用条件が入力されると、運用条件と、運用状態格納部に格納された複数の評価指標の値と、運用想定値格納部に格納された複数の評価指標の値と、に基づいて、実行される文字列の認識の結果を評価する複数の評価指標の値が運用条件を満たす値に近づくように、複数の文字列認識部のいずれかを選択する。【選択図】図１