PROBLEM TO BE SOLVED: To provide an image processor capable of accurately and clearly distinguishing a normal region from an abnormal region.SOLUTION: A result correction part 120 receives from an affected area detection part 110 respective detection results of a target still image and a reference still image and uses the detection results to identify a position, size, and range of an abnormal region. Then, a reference still image in which result R(t) of reference detection overlaps one-fourth or more of a result R(0) of a target detection is detected. The number of pixels in an overlapping region is used as reference weight G(t). For the reference still image in which the result R(t) does not overlap one-fourth or more of the result R(0), the reference weight G(t) is 0. Next, normalized weight g(t) is calculated. Values obtained by multiplying a reference feature quantity P(t) by the normalized weight g(t) is obtained for all the reference featured value P(t), a target feature quantity P(0) is added thereto, and an additive feature quantity S(0) is obtained. The additive feature quantity S(0) is used to detect the abnormal region in the target still image.COPYRIGHT: (C)2013,JPO&INPIT【課題】正常部と異常部とを正確かつ明確に分別することが可能な画像処理装置を得る。【解決手段】結果補正部120は、病変検出部110から対象静止画像及び参考静止画像の各検出結果を受信し、これらの検出結果を用いて、異常部の位置、大きさ、及び範囲を特定する。そして、参考検出結果R(t)が対象検出結果R(0)と1/4以上重複する参考静止画像を検出する。そして、重複する領域の画素数を参考重みG(t)とする。1/4以上重複しない参考静止画像に関しては、参考重みG(t)は0となる。次に、正規化重みg(t)を算出する。参考特徴量P(t)に正規化重みg(t)を乗じたものを全ての参考特徴量P(t)に関して求め、対象特徴量P(0)に加え、加算特徴量S(0)を求める。加算特徴量S(0)を用いて、対象静止画像における異常部を検出する。【選択図】図1
