GB2286706A

GB2286706A - Object recognition process

Info

Publication number: GB2286706A
Application number: GB9505355A
Authority: GB
Inventors: Olga Olegovna Verovenko
Original assignee: Individual
Current assignee: Individual
Priority date: 1993-07-16
Filing date: 1994-06-30
Publication date: 1995-08-23
Also published as: GB2286706A8; GB2286706A9; FR2708362A1; DE4495111T1; RU2037203C1; WO1995002868A1; GB9505355D0; DE4495111C2

Abstract

The invention relates to information technology, in particular to object recognition, and can be used to construct recognition systems based on modern computers, for modelling recognition processes which use modern computers, and also for training text editors and in language teaching. In the proposed object recognition process, a system of attributes is created corresponding to the object to be recognized, the recognition system reads the data file submitted for processing, the attributes corresponding to the objects contained in the data file are identified and compared with the created systems of attributes, and the objects possessing the created attribute systems are identified. The novelty of the invention lies in the fact that the creation of a system of attributes corresponding to each object to be recognized is done by reading in sequence the samples from a first group of samples, each of which contains models of the object to be recognized, the attributes corresponding to the object to be recognized are identified, the attributes identified in each model of the object to be recognized are compared among themselves, the samples being arranged in order of ascending complexity of the model of the object to be recognized included within them; after each sample from the first group has been processed, the sample from the second (control) sample group containing models of the same object to be recognized is read on the basis of the established characterizing attributes.

Description

METHOD FOR OBJECT RECOGNITION The present invention relates to the information technology, namely to the field of the object recognition (identification), and can be used for making recognition systems with employment of modern computers, for simulating processes with employment of modern computers, and for training text editors, for teaching foreing languages etc.

Methods for the object recognition recently become wider spreaded for solving a number of scientifical and practical problems. Methods for the object recognition are known characterized by the next set of similar features when during the object recognition the following steps are performed: genereting the determined attribute sets corresponding to the object to be recognized, comparing object attributes presented for the recognition with predetermined attribute sets, and deciding about the object to be recognized on the basis of said step of comparing (Zdor S.E., Shirokov V.B. Optical search and recognition. - Moscow: "Nauka", 1973, pp.l35-141). The difference of one method for the object recognition from another consists substantially in using various ways for classifying objects. Thus the metric approach is based on generating references and sampling objects, said sampling on one hand must not result in redundant description and on other hand must reproduce the original object as close as possible during its comparison with a reference.

The structural approach to the object recognition is based on generating some set of elementary attributes. Such attributes as "angle", "cross", etc are used for these elementary attributes.

The disadvantage of that methods is, first, in that they have a limited area of utilization because in some cases none of determined attribute sets corresponding to objects to be recognized are presented, for example in differential disease diagnostics. Second, the effectiveness of the information processing by known methods is not great because during the recognition of each object it is necessary to look through all predetermined attribute sets and to select those one corresponding to the object to be recognized.

The present invention is based on a problem of performing a method for the object recognition with such a sequence of operations for processing of main and addition information blocks, which would provide the generation of an attribute set for the object classification directly during the processing of addition information blocks, thereby allowing to rise efficiency of the processing of large information blocks including objects related to several complex information groups.

The proposed method for the object recognition, as the known ones, comprises the steps of: generating the system of attributes corresponding to the objects to be recognized, reading by a recognition system an information array presented for processing, selecting object attributes included in said information array, comparing selected attributes of objects to be recognized with generated systems of attributes, and selecting objects having generated systems of attributes.

In addition to above steps the method according to the present invention comprises operations characterizing the step of generating two sample clusters and also characterizing the gist of one of operation comprised in known methods for the object recognition, namely, the step of generating the system of attributes corresponding to the object to be recognized.

According to the present invention the step of generating systems of attributes corresponding to each object to be recognized comprises the steps of: reading the first cluster of samples each having examples of said object to be recognized, selecting attributes of said object to be recognized, comparing attributes selected in each example of said object to be recognized one to another, selecting common attributes of said object to be recognized, said first cluster samples being ordered with an increasing complexity of examples of said object to be recognized comprized therein, and after processing each sample of said first cluster, reading a sample from the second cluster of samples, said sample having examples of said object to be recognized, and then recognizing the object of interest on the basis of previously generated common attributes, a complexity of examples of each sample in said second cluster of samples being to correspond to a complexity of examples comprised in a preceded sample of said first cluster.

The advantage of the proposed method for the object recognition in accordance with the present invention is in the fact that as a result of analysing a finite sequence of first cluster samples including examples of the object to be recognized, the recognition system generates its own version of the system of attributes corresponding to each particular object to be recognized, in other words, an adaptation of a vocabulary and a logic of the recognition system is performed to recognize a given object. This fact allows to use the proposed method to recognize such objects which attributes are not predetermined, and therewith said method can be used as appropriately both in recognition systems based on modern computers and in an operator training.

A drawing shows a block diagram of the recognition system carrying out the proposed method, in which system an information carrier is supplied to an input of a reading unit 1.

An output of the reading unit 1 is coupled to an input 2 of an analyser 3 which output being connected directly to a registration unit 4 and, via a serial connection of a logic unit 6 and a program unit 7, to its own inputs 5. The logic unit 6 is coupled also to a memory uniy 8.

The method for the object recognition is performed as follows.

Previously the first cluster of samples is generated each sample including examples of an object to be recognized. The object to be recognized must be emphasized in any way in each example included in said sample. A technique of emphasis of the object to be recognized depends upon construction particularities of the information reading unit and, in principle, any of known techniques of emphasis of the required information content can be used. Also, the second cluster of (checking) samples is generated each checking sample including examples of the same object to be recognized but an information related to the object to be recognized in that examples does not emphasized specially. An information related to each sample is recorded on a corresponding carrier.

Then carriers corresponding to first and second cluster samples are ordered by complexity increasing of the information recorded therein. Thus the preliminary stage for performing the method is finished.

Then the carrier with the recorded first sample from the first cluster of samples is supplied to an input of the reading unit 1 and the step of reading an information recorded on that carrier is performed. From the output of the reading unit 1 the information comprised in the first sample supplies in the form of sequence of e.g. electrical signals to an input 2 of the analyzer 3. By means of analyzer 3 the information in each example included in the first sample is processed. The step of processing examples included in the first sample can be performed either in serial or in parallel depending on the technical possibilities of the analyzer 3. In each example of the object to be recognized the specific elementary attributes are selected including their relative positions and positions of the object to be recognized relative to other objects included into a corresponding example. Attributes selected in each example are compared one to another by the logic unit 6 and as a result said attributes are divided into clusters of "similar" or identical attributes. Quantity of selected attribute clusters will determine a first version of the system of attributes corresponing to the object to be recognized.

Thus a step of selecting the subset of attributes corresponding (as a first approximation) to the object to be recognized from a set of attributes fed into the analyzer 3 circuit by the logic unit 6 is performed. An adjusting of the analyzer 3 to classify objects in accordance with the generated system of recognition attributes is performed by the program unit 7 to which input the signals from the output of the logic unit 6 are supplied. At the same time this version of the generated attribute system is stored in the memory unit 8.

To check the results of the "teaching"of the recognition system to detect a given object the carrier with a recorded first sample from the second (checking) sample cluster is supplied to the input of the reading unit I. After that the following steps are performed: selecting attributes of all objects contained in each example of the analyzed checking sample and comparing the attnbutes presented for the object recognition with the generated system of attributes. Based on this step of comparing an object having the generated system of attributes is selected. Recognition results come to the registration unit 4. If a recognition object validity index derived from the processing of the first checking sample would be not lower than a needed value then the carrier with a recorded second sample from the first sample cluster having examples of the object to be recognized is supplied to the input of the reading unit 1. At that time outputs of the logic unit 6 connected to the program unit 7 are reset and the analyzer 3 is switched into a mode corresponding to the mode of processing the first sample of the same sample cluster.

Then the information included in the second sample is processed in a sequence described above and a subset of attributes corresponding the object to be recognized is derived. In principle two cases are possible here: the derived subset of attributes either coincides or partly not coincides the subset of attributes derived during the processing of the first teaching sample.

In the first case carriers with recorded checking samples are supplied to the input of the reading unit 1. If during the processing of some next checking sample the index value of a validity of an object to be recognized would be derived less than a required value then a carrier with a recorded sample from the first sample cluster is supplied to the input of the reading unit I, the number of that teaching sample being the same with the number of the checking sample while processing the latter the understate validity index of the object recognition was derived. As a result a refined subset of attributes corresponding to the object to be recognized is derived, i.e. a situation qualified above as a second case take place.

In the second case an extended attribute subset comprising attributes of subsets derived during the processing of both samples of the first sample cluster is formed by the logic unit 6. Then a carrier is supplied to the input of the reading unit 1, with a recorded second checking sample processed by the analyzer 3 adjusted onto the extended attribute subset. A whole number of teaching samples and corresponding checking samples is determined from the complexity of the recognition problem and depends on a source of the prior information about the object to be recognized.

Claims

A method for object recognition comprising steps of: generating the system of attributes corresponding to objects to be recognized, reading by a recognition system an information array presented for processing, selecting object attributes included in said information array, comparing selected attributes of objects to be recognized with generated systems of attributes, and selecting objects having said generated systems of attributes, c h a r a t e r i z e d in that the method further comprises a step of generating two sample clusters, and said step of generating systems of attributes corresponding to each object to be recognized comprises steps of: reading the first cluster of samples each having examples of said object to be recognized, selecting attributes of said object to be recognized, comparing attributes selected in each example of said object to be recognized one to another, selecting common attributes of said object to be recognized, said first cluster samples being ordered with an increasing complexity of examples of said object to be recognized comprised therein, and after processing each sample of said first cluster, reading a sample from the second cluster of samples, said sample having examples of the same object to be recognized, and then recognizing the object of interest on the basis of previously generated common attributes, a complexity of examples of each sample in said second cluster of samples being to correspond to the complexity of examples comprised in a preceded sample of said first cluster.