Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1998-05-19
2002-07-23
Couso, Jose L. (Department: 2721)
Image analysis
Pattern recognition
Feature extraction
C382S110000, C382S154000, C382S165000, C356S601000, C250S559190
Reexamination Certificate
active
06424745
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to the field of optical object recognition, and more particularly to the use of local radii of curvatures to identify products and materials.
BACKGROUND OF THE INVENTION
Bar-code readers are used extensively in the retail grocery industry at checkout stations to identify tagged items affixed with bar-code tags. An item is identified by means of its bar-code using a data base stored in a host computer. Typically, a description of the item and its price are printed on a grocery receipt and an ongoing price total is kept as additional items are scanned. The use of bar-code readers has generally been well received by the public, due in part, to the reliability and reduced time spent in the checkout line. However, a reliable system is needed to identify items for which it is undesirable to attach bar-code labels, for example, fresh produce, such as fruits and vegetables.
The use of optical devices to characterize and recognize fresh produce has been explored to some extent in the prior art. Some general considerations involved in the optical identification of these products are discussed in a number of publications, including for example, “Postharvest Handling: A Systems Approach”, by R. L. Shewfelt and S. E. Prussia (Academic Press, New York). An example of a particular application of produce recognition is a study by L. Bochereau et al. (J. Agric. Eng. Res. (1992) Vol. 51, 207-216) showing that near infrared spectra can be used to determine the quality of golden delicious apples, provided that a careful neural-network analysis is performed in the wavelength range from 1,000 to 1,500 nm.
Specifically, a number of devices are disclosed in the prior art for recognizing products using image processing techniques. For example, U.S. Pat. No. 5,546,475 issued to Bolle et al. and assigned to IBM Corp., discloses a device that uses image processing to recognize objects within a scene. The prior art image processing device takes a first image at a high illumination level and a second image at a lower illumination level by using a controlled illumination source. A segmentation routine separates the background image from the object image by comparing the two images. Feature extraction is performed on the processed image and compared against stored reference images. In one embodiment, shape is used as a recognition feature, where local boundary curvatures are used for the shape measurements.
A disadvantage of this prior art device is that direct curvature information is not available since the device illuminates and captures images of the entire object. The curvature information is then derived from the available grey scale information. Since the grey scale information is directly related to the intensity levels captured by the camera, color variations in the objects make it difficult to determine accurate and consistent curvature information. A further drawback is that the intensity is also affected by the orientation of the object. As a result, an object recognition system is required that is independent of the varying characteristics of the object.
SUMMARY OF THE INVENTION
The present invention is a method and a system that increases the accuracy and reliability of object recognition by analyzing distributions of local radii of curvatures to determine the unknown object. The system provides optical recognition of products that do not readily lend themselves to standard labeling techniques. Importantly, the system can be employed with, and is integratable with, retail checkout systems which use bar code reader technology. The resulting combination provides a system which processes both labeled and non-labeled products reliably and accurately.
In an exemplary embodiment of the present invention, a surface portion of an object is illuminated with a pattern of light that permits the extraction of three dimensional coordinates for a set of points on the surface portion of the object. An image data set of the surface portion of the object is then captured with a capture device that is positioned at an angular offset with respect to a source of the light. That is, the combination of the light pattern and the imaging device together generate a two dimensional captured image, from which it is possible to extract the three dimensional coordinates for the set of points on the surface portion of the object. An exemplary embodiment of such a light pattern may consist of a series of separately identifiable arcs projected on the surface portion of the object. A set of local radii of curvatures are then determined for a set of selected data points in the image data set. A spectrum representing a distribution of the curvatures is then computed for the set of local radii of curvatures. If the image data set is for the generation of a library of spectra, it is processed with other spectra of curvatures to form a reference library of known objects. If the image data set is for an unknown object, then it is compared against the reference library of known objects to make an identification.
Advantageously, the method of the present invention provides three-dimensional curvature measurements because the angle between the light source and the capture device has the effect of making the reflected patterned light look curved. Moreover, the method is independent of object variations and object orientation since the library of spectra generally includes curvature distributions reflecting potential orientations of the objects. The amount of computer memory required is not significantly increased because basis functions and basis coefficients permit low dimensional representation and expansion of the spectra. Importantly, the present invention has to maintain only a fraction of the original data to make the required identification.
The above factors make the present invention product identifier an advantageous and practical solution for use at supermarket checkout counters to identify fresh produce, such as fruits and vegetables, which are not readily identifiable with a bar-code label. Moreover, the product to be identified could be sitting on a scale and its weight could be measured during the identification process. In a similar manner to grocery items containing bar-code labels, the name and price of the item would be printed on a grocery receipt using the item identification, along with the weight and the price per pound which had been previously stored.
REFERENCES:
patent: 4873651 (1989-10-01), Raviv
patent: 5060290 (1991-10-01), Kelly et al.
patent: 5085325 (1992-02-01), James et al.
patent: 5164795 (1992-11-01), Conway
patent: 5546475 (1996-08-01), Bolle et al.
patent: 5548405 (1996-08-01), Motosugi
patent: 5729704 (1998-03-01), Stone et al.
Hansen Mark Henry
Hess Harald F.
Mitra Partha Pratim
Thomas Gordon Albert
Couso Jose L.
Desire Gregory
Gibbons Del Deo Dolan Griffinger & Vecchione
Lucent Technologies - Inc.
LandOfFree
Method and apparatus for object 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 and apparatus for object recognition, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for object recognition will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2887860