Scene feature values are computed from a scene of an R image of an endoscope image. Determination analysis is performed using a statistical or nonstatistical discriminator, with the scene feature values used as feature value vectors. The scene feature values are classified into a plurality of feature value classes (e.g., a class of lumen dark sections and a class of folds) closely relating to an endoscope insertion direction. An essential feature value class is computed, an insertion direction corresponding to a feature value in the essential feature value class is computed, and a high-accuracy insertion direction marker and the like are displayed. With the above-described operations, an endoscope insertion direction detecting device and an endoscope insertion direction detecting method capable of detecting an insertion direction with high accuracy even in the case of a scene in which a plurality of feature values are present are realized.
