IMAGE FEATURE EXTRACTING DEVICE, IMAGE RECOGNIZING METHOD AND IMAGE RECOGNIZING DEVICE

摘要

PURPOSE:To provide the high recognition rate in the image recognizing device using a neural network. CONSTITUTION:A bit image prepared by an input part 1 is processed by the neural network constituting a feature extraction part 2 and a recognition part 3, and an image is recognized. First of all, the feature extraction part extracts a line segment in a specified direction or a specified geometrical shape from the bit image by a first neural network constituting a shape extraction part 4. In a second neural network constituting a direction detection part 5, the direction and amount of letting a second specified shape exist is extracted corresponding to a first specified shape. Based on this feature, the image is recognized by a third neural network constituting the recognition part.