Image analysis – Pattern recognition – Context analysis or word recognition
Patent
1995-12-01
1998-06-30
Moore, David K.
Image analysis
Pattern recognition
Context analysis or word recognition
382190, 395 26, 395 263, G06K 972, G06K 946, G10L 506
Patent
active
057745870
DESCRIPTION:
BRIEF SUMMARY
The present invention relates to methods of recognition of various objects, for example graphic images, texts, oral reports, radar signals, etc.
BACKGROUND TO THE INVENTION
The problem of recognizing an object is solved in general by means of comparing this object or a part thereof to certain predetermined reference sample. To facilitate this comparison, the object under recognition and reference sample may be replaced with their representations, for example in the from of code signals.
Evident examples of the recognition of objects are the selection of an appropriate element in a puzzle for children in order to arrange the necessary pattern, or the identification of a well-known melody by a small part thereof. But such tasks have no structure which can be reduced to simple algorithmic solutions.
One of the possible ways to carry out the problem of recognition is to break down the object to its components (dots of all image, frequency harmonics of a vocal signal) and to compare the set of these components with reference sets of similar elements. But technical and time limitations, e.g. a large volume of equipment (memory) in the case of a parallel comparison to many references, or a long time necessary for such comparison in the case of the subsequent comparing with references, do not make it possible to use such a method for the recognition of complicated objects.
Usually "element by element" comparison of the object is replaced by the comparison only of its most characteristic features: predetermined pattern on a drawing, the of eyes on a portrait, the sets phonemes of a vocal signal (See U.S. Pat. No. 5,016,282, C1. G 06 K 9/900, published in 1991; U.S. Pat. No. 5,195,166, C1. G 10 L 9/00, published on 16.03.93). But if such distinctive features are absent in the part of the object to be subjected to recognition, there will be no recognition even if this part of the object is the part of a reference object.
Another method to replace this "element by element" comparison is the selection of certain primitive elements which if combined make it possible to represent any reference object (for example, combination of seven basic strokes and their modifications in Chinese hieroglyphs). In this case recognition of the objects reduces to identification of primitive elements in these objects and to the definition of their combination (international application WO 89/05494 C1. G 06 K 9/22, 9/78, published in 1989). This method is intended for the recognition of the characters in texts and cannot be used for other objects. In addition the presence of various different combinations, consisting of a limited amount of primitive elements is hardly more simple than the method of the "element by element" recognition.
In U.S. Pat. No. 5,133,021 C1. G 06 K 9/00, published in 1992 another possible recognition method based on the theory of adaptive resonance is described. According to this method an input signal which can be represented either as a dimensional or as time sample is passed though an adaptive filter and chooses one of the patterns. Then the deviation between the input signal and the chosen pattern is defined. If this deviation is larger than the predetermined one of the chosen pattern is rejected, and otherwise--this deviation is being adapted to the maximum possible extent to the input signal. The necessity of such an adaptation may be the source of errors in the case of a small number of patterns.
The above mentioned drawbacks can be avoided if the method of associative recognition based on the use of one-dimensional non-linear dynamic systems of a type of mapping of a segment into itself is used (Yu. V. Andreev, A. S. Dmitriev, S. O. Starkov. Images Processing on the basis of one-dimensional dynamic systems. Software Complex "Informchaos". Preprint N.2 (584). Institute of Radioengineering & Electronics RAS, Moscow, 1993 pg. 4-36). The above mentioned dynamic system is formed in such a way that it contains various limit cycles each corresponding to one of the references. If such a dynamic system, in the ca
REFERENCES:
patent: 4348553 (1982-09-01), Baker et al.
patent: 4955056 (1990-09-01), Stentiford
patent: 5179542 (1993-01-01), Reese et al.
patent: 5287275 (1994-02-01), Kimura
patent: 5446828 (1995-08-01), Woodall
Andaeev Yury
Belsky Yury
Dmitriev Alexander
Kuminov Dmitry
Panas Andzei
Chen Wenpeng
Moore David K.
LandOfFree
Method of objects recognition 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 of objects recognition, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of objects recognition will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-1868794