Image analysis – Pattern recognition – Template matching
Reexamination Certificate
2001-03-07
2004-06-15
Johnson, Timothy M. (Department: 2625)
Image analysis
Pattern recognition
Template matching
C382S181000, C382S309000, C348S254000
Reexamination Certificate
active
06751353
ABSTRACT:
FIELD AND BACKGROUND OF THE INVENTION
The invention relates to the information image recognition and can be used for adaptive processing and classification of information flow.
At present, a variety of adaptive image recognition techniques are known for both pictures and other forms of images such as sound or tactile ones.
In the paper “Research decision-making system by complex-image fragments using neuron-like algorithms” (Izv. Vyssh. Uchebn. Zaved., Radiofiz., vol. 37, 1994, No. 8, pp. 961-986 (in Russian)) by V. G. Yakhno et al. is described an approach which outlines the following ways for solving the problem of information image adaptive recognition: signal encoding; back recovering of signals from their code descriptions; forming a set of estimates for code descriptions and recovered signals; forming a tree of possible solutions; determining the efficiency or correctness of encoding; recovering and estimating; and realigning operating algorithms in the case of their inefficiency.
In this approach, as mentioned above, only general ways are outlined, but there are no particular operation sequence for adaptive image recognition and no devices for implementing that approach. The recognition algorithms were assumed to be predetermined in that paper.
In the RF Patent No. 2090928 (Int. C1. G 06 K 9/00, Sep. 20, 1997), is described a method for adaptive signal recognition on the state of object, which method includes steps of creating recognition standards and correcting them during a training, comparing these standards with a signal being analyzing, forming estimates of object states, and comparing these estimates with the predetermined recognition threshold, the step of correcting the formed standards being iterative. A drawback of this method is that the recognition is performed by comparing the signal being recognized with the standards obtained at the training stage and not liable to correction during recognition, i.e., obtained without checking the degree of actual correspondence between these standards and the signal being recognized. Thus, a situation can arise in which the signal being recognized is different from the signals by which the standards were formed and corrected at the training stage, which results in recognition errors.
The U.S. Pat. No. 4,491,962 (Int. C1. G 06 K 9/00, Jan. 1, 1985) describes a recognition method which determines the optimal threshold for obtaining a binary image signal. In this method, the video signal of an object is converted to a binary image signal on the basis of each of various values of the threshold level, in which determining a template expected to lie somewhere on the object surface is chosen as a standard one. Then a set of partial templates is cut out of the given region of a binary image signal corresponding to each value of the threshold level, and the degree of coincidence between each partial template and the indicated standard template is determined. The value of the threshold level is chosen which corresponds, as a whole, to the maximum degree of coincidence between this set of partial templates and the standard template taking into account the coincidence degrees for each value of the threshold levels. The object image is divided into parts, and the maximum degree of coincidence between the set of partial templates and the standard template is chosen. This process is performed for all parts of the image, and the final sum of maximum values of all surface parts is used as the coincidence degree in whole. A drawback of this method is that the image which can be absent in the picture being recognized and is not liable to correction can be used as the standard template. This can lead to incorrect recognition.
This method is implemented in a system for adaptive recognition of information images described in the same U.S. Pat. No. 4,491,962 and comprising an information image adaptive processing unit intended for processing an, input information image in accordance with the algorithm developed beforehand for recognizing information images; a memory unit intended for storing the information image processed by the information image adaptive processing unit at least until finishing the image recognition procedure; a comparison unit intended for comparing the input information image being processed in the information image adaptive processing unit with the model image and for estimating discrepancies between the compared images; and a decision unit intended for deciding on the recognition of the information image. The drawbacks of this system are the same as those of the method being implemented by this system.
The U.S. Pat. No. 5,054,093 (Int. C1. G 06 K 9/00, Oct. 1, 1991) describes a parallel multiunit adaptive nonlinear coupler and classifier for implementing another method for an adaptive recognition which is closest to the claimed method. This known method comprises steps of: forming in advance models of probable information images (pattern prototypes related to different classes of patterns); processing each input information image in accordance with the algorithm developed in advance for recognizing information images; comparing the input information image being processed (input pattern) with the prototype; and correcting the recognition algorithm by storing a new prototype if the input pattern does not coincide with a region of the initial prototype; and, in the case of coincidence, modifying the prototype in the memory for matching it with the input pattern. A drawback of this method is that only one recognition algorithm with replaceable prototypes is used, which narrows the applicability of this method. Moreover, in this method, a coincidence is estimated with the prototypes of the information images already processed, which precludes to check the correctness of the processing procedure itself and its effect in the result being obtained.
As was mentioned above, this method is implemented in an apparatus which is chosen as the nearest analogue of the claimed one and comprises: an information image adaptive processing unit intended for processing the input information image in accordance with the algorithm developed in advance for recognizing information images; a comparison unit intended for comparing the input information image being processed in the information image adaptive processing unit with the model image, and for estimating discrepancies between the compared images; a model pattern correcting unit intended for correcting the pattern image by signals from the comparison unit. The drawbacks of this system are the same as those of the method being implemented by this system.
SUMMARY OF THE INVENTION
The task of this invention is to develop a method and a system for adaptive recognition of information images which could optimize the procedure for choosing the most exact and fast algorithms for processing a large flow of information in accordance with the type of the problem to be solved and the nature of the data being processed. In other words, this invention is aimed at solving the problem of accelerating a search for adequate processing and decision algorithms to make such work automatic. At the first stage, the system in which this invention has been implemented is necessary to develop and create automatic recognizing devices, and then can itself be included as a component part into an adaptive self-tuned recognizing system which is automatically aligned to changed operating conditions.
Thus, the technical result being reached is in increasing a recognition accuracy for the wider class of objects.
To solve this problem and reach the required technical result, a method is claimed for adaptive recognition of information images, comprising steps of: processing each input information image in accordance with a recognition algorithm; comparing the input information image being processed with a model image in order to estimate discrepancies between the compared images; comparing at least one estimate of the obtained discrepancies with at least one predetermined threshold; when said estimates exceed the predetermi
Alexandrovich Sborshikov Vladimir
Alexandrovich Telnykh Alexander
Grigorievich Yakhno Vladimir
Ivanovich Khil'ko Alexander
Nikolaevich Bondarenko Boris
Choobin Barry
Notaro & Michalos P.C.
LandOfFree
Method for adaptive recognition of information images and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method for adaptive recognition of information images and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for adaptive recognition of information images and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3308973