Image analysis – Learning systems – Trainable classifiers or pattern recognizers
Reexamination Certificate
2000-02-17
2003-11-11
Ahmed, Samir (Department: 2623)
Image analysis
Learning systems
Trainable classifiers or pattern recognizers
C382S190000
Reexamination Certificate
active
06647139
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to an object recognition method for recognizing an object from the image obtained from image input means, apparatus of the same and a recording medium in which the program of this method is recorded.
BACKGROUND OF THE INVENTION
Object recognition methods hitherto known include a feature extraction using Karhunen-Loeve transform, and similar methods. For example, “Visual Learning and Recognition of 3-D Objects from Appearance” by H. Murase and S. K. Nayer (International Journal of Computer Vision, 14, 1995), Japanese Laid-open Patent No. 8-271223, and Japanese Laid-open Patent No. 9-53915 are known.
A conventional object recognition apparatus is explained by referring to a drawing. In
FIG. 22
, a conventional object recognition apparatus comprises an image input unit
11
such as a camera for entering an image, a learning model memory unit
13
for preparing and storing local models of target object for recognitions from learning images, a feature extractor
12
for extracting the feature of an input image, a learning feature memory unit
14
for storing the feature (learning feature) of the model, a matching processor
15
for matching the feature of the input image with the feature of each model, and an object type estimator
16
for judging and issuing the type of the target object for recognition in the input image. Herein, the type refers to the individual or the kind.
The operation is described below. When an input image including a target object for recognition is entered in the feature extractor
12
through the image input unit
11
, the feature extractor
12
extracts a feature from the input image, and issues the feature to the matching processor
15
. The matching processor
15
sequentially searches the models from the learning model memory unit
13
, and selects the learning feature from the learning feature memory unit
14
. The similarity measure between the input image feature and the learning feature is calculated, and is issued to the object type estimator
16
. Thus, the matching processor
15
repeats the procedure of similarity measure calculation and output by using the model of the learning model memory unit
13
. When the similarity measure is the maximum, the object type estimator
16
determines to which type of models the target object for recognition included in the input image belongs.
The input image is overlapped with various learning images, and the overlapping degree is judged by using the similarity measure, and therefore the object equal to the learning image can be recognized, but when an object not being learned is included in the input image, it is difficult to estimate and recognize the object.
Or when recognizing the same object as the learning image, it was difficult to recognize if there is no information about the distance to the existing position of the object. To obtain the distance information by the imaging device only, a three-dimensional camera is needed, but the signal processing is complicated.
SUMMARY OF THE INVENTION
It is hence the object of the invention to present a method of recognizing an object accurately.
The object recognition method of the invention is an object recognition method comprising at least a learning step of learning a first entered image, and a recognition step of recognizing an entered second image, in which the learning step includes a step of entering the first image including the object to be learned, a step of dividing the entered image into a first partial image, a step of classifying the first partial image into plural classes, a step of calculating the feature extraction matrix in every classified class, a step of calculating a first feature by using the feature extraction matrix from the partial image classified in each class, and a step of storing the data of the first feature, and the recognition step includes a step of receiving a second image including the object to be recognized, a step of dividing the entered image into a second partial image, a step of calculating a second feature by using the feature extraction matrix from the second partial image, a step of calculating the similarity measure of the both by using the stored first feature data and second feature, a step of recognizing the object in the second image by using the similarity, and a step of issuing the result of recognition.
In this constitution, by setting a step of classifying the image to be learned in particular, the recognition method of this invention can recognize the object at high accuracy than in the prior art.
REFERENCES:
patent: 5842194 (1998-11-01), Arbuckle
Kunii Mihoko
Nagao Kenji
Ahmed Samir
Matsushita Electric - Industrial Co., Ltd.
RatnerPrestia
LandOfFree
Method of object recognition, apparatus of the same 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 of object recognition, apparatus of the same and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method of object recognition, apparatus of the same and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3184637