An image classifier receives input images and assigns each input image to one of a plurality of image classes. The image classifier includes plural class distribution maps, each based on a plurality of features evaluated on training images, and each representing those feature values that occur at least once among the training images belonging to the corresponding class. The image classifier further includes means for constructing a test map by evaluating the plurality of features on the input image. The image classifier further includes means for comparing the test map to the class distribution maps in order to identify which one of the class distribution maps has the least distance to the test map. At least one of the features is defined according to a rule that relates to the shapes of images of at least one image class. One advantageous method for evaluating features is carried out storing the input pattern, in a digital memory, as a point in a representational space. A digital data-processing device retrieves one or more line segments from a library of line segments constructed in the representational space, wherein each said line segment is subdivided into plural indexed sub-intervals and each line segment is associated with a respective feature. The digital data-processing device projects the input pattern onto each of the retrieved line segments. For each said line segment, the feature associated with that line segment is assigned a value equal to the index of the sub-interval onto which the input pattern is projected.

A method for classifying character line patterns in a dictionary, includes the steps of: (a) obtaining a first class to which one or a plurality of first line patterns belong; (b) obtaining a second class to which one or a plurality of second line patterns belong, each of the first and second line patterns being characterized by structural information, (c) determining whether or not the structural information of the first line pattern and the structural information of the second line pattern are similar to each other in accordance with a predetermined rule; and (d) combining the first and second classes with each other when the step (c) determines that the structural information of the first line pattern and the structural information of the second line pattern are similar to each other, so that a new class is obtained, to which the first and second line patterns belong.

Image scanning process for detecting set forms, the main characteristic of the process being that it includes a first stage consisting in analysing signals representing points on the said image, in selecting first points defining areas in which the variation in contrast exceeds a given threshold, and in determining the direction of the tangent line relative to each of the said first points; a second stage consisting in detecting whether, for each of the said first points there exist other first points lying on at least one line having a set length and angle in relation to the direction of the line tangent with the point in question and in projecting onto said tangent line if the direction of the said other point lying on the said line has a second set angle in relation to said tangent line; a third stage consisting in memorizing and accumulating the number of the said projections relative to the said projection points and in extracting the said projection points whose accumulated value exceeds a given number threshold; and a fourth stage consisting in selecting from among the said extracted points those defining the set form.

A method for recognizing previously localized characters present in digital gray tone images, particularly for recognizing characters struck in metal surfaces, whereby, for training a trainable character recognition routine, steps are provided to generate reference characters presented line-like and to deposit these reference characters in a working memory of the trainable character recognition routine, whereby the number and nature of the reference characters correspond to the character set from which characters are to be recognized. For recognizing characters, steps are provided: to read the digitized character of the localized character to be recognized into a character recognition routine and an appertaining gray tone image is provided, to pre-process the character to be recognized so that a classification of the appertaining character can be implemented; to compare the preprocessed character to all reference characters previously learned by the character recognition routine, to implement a majority decision for identifying that reference character that has the greatest plurality of sub-features coinciding with the character to be recognized, and to produce a result signal from the character recognition routine for further processing thereof.

An apparatus for recognizing visual patterns of handprinted characters or the like by means of an optical character reader which reads the patterns by a so-called outermost point method. While tracing the contour of the pattern stored in a two-dimensional memory, the distances from the starting point of the tracing and the integrated values of the coordinates of the points traced on the contour from the starting points are simultaneously obtained successively to extract outermost points for the series of contours. According to this outermost points, the contours are segmented into the convex line segments, concavity line segments and hole segments and the corresponding parameters of features of each segment are detected. Simultaneously, the convex line segment of which is shorter than a predetermined length is rejected and the remaining segments are subjected to matching operation in accordance with preliminarily prepared dictionary, thereby making a decision as to the pattern's identity.