Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
2000-01-18
2003-11-25
Mariam, Daniel G. (Department: 2723)
Image analysis
Pattern recognition
Feature extraction
C382S195000, C382S203000, C382S218000, C382S283000, C382S316000
Reexamination Certificate
active
06654497
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to an image processing apparatus, method and storage medium.
2. Related Background Art
Conventional contour extraction methods include (1) a binarizing method of binarizing an image and tracing a boundary of an obtained binary pattern to extract a contour, (2) a tracing method of using a multivalued edge image, referring to the past tracing direction and tracing a local maximum edge pixel to extract the contour, (3) an area limiting method of detecting a boundary point of an object and a background in a pre-limited area to extract the contour, and (4) an active contour method of moving a closed curve surrounding an object to be cut out so that the predefined energy is minimized to extract the contour of the object.
The following method is known as the binarizing method. Specifically, a method described in Japanese Patent Publication No. 7-62873 comprises the steps of first binarizing an input image with an appropriate threshold value, removing noise from the obtained binary pattern, using an attended contour point as a center to search for the pixel in the vicinity in a clockwise or counterclockwise direction, detecting the next contour point, using the found contour point as a new attended point, and subsequently repeating the same processing to extract a contour pixel. In a method described in Japanese Patent Application Laid-Open No. 8-50655, an edge image constituted of an edge with one pixel width is first prepared. Specifically, a differential space filter or the like is applied to the input image to sharpen the image. If the edge width does not reach one pixel, a fine line processing is applied. After the edge image is prepared, in the same manner as in the method described in the Japanese Patent Publication No. 7-62873, the pixel in the vicinity is searched for in the clockwise or counterclockwise direction centering on the attended point to detect the next edge image.
In the tracing method, the multivalued edge image is used, and the pixel with the maximum edge strength is selected as the contour pixel from the pixels in the vicinity of the attended pixel. In the invention of Japanese Patent No. 2645995, on the assumption that it is natural to perform the next tracing in the same direction as the past tracing direction, the pixel edge strength along the past tracing direction is strongly weighted, the edge is traced, and the contour line is determined from a trace result.
In the area limiting method, the contour line is extracted only from a strip area which is set so as to include the contour line. In a method described in Japanese Patent Application Laid-Open No. 63-55101, an operator first roughly traces along the contour line. The strip area having a predetermined width is selected along the traced locus, and the image contour is determined from a tone value difference in the strip area. A method described in U.S. Pat. No. 5,617,487 (Japanese Patent Application Laid-Open No. 7-92650) comprises the steps of connecting at least two points designated on the contour line to form the strip area, and repeatedly executing a processing of detecting the contour point of the object in the width direction of the strip area while moving in the longitudinal direction of the strip area, to extract the contour line of the object.
In the active contour method, the contour line of the object is extracted by moving the closed curve surrounding the object to be cut out so that the predefined energy is minimized. For example, in “Snakes: Active Contour Models” authored by M. Kass, A. Witkin and D. Terzopoulos, International Journal of Computer Vision, Vol. 1, No. 4, pp. 321-331, 1988 is described a method of contracting/deforming the contour so that the sum of an energy applied in accordance with the shape of a contour model (internal energy), an energy applied in accordance with an image property (image energy), and an energy applied from the outside (external energy) is minimized.
The binarizing method has the following problems. Specifically, since the binarizing method is limited to an image from which the object can be extracted by a binarizing processing, for example, an image with a plain background, the method cannot be applied to a natural image. Even if the background is plain, in the binarizing processing of the natural image photographed under lighting conditions which cannot be managed, it is difficult to set the threshold value, and it is also difficult to automate the processing. After the binarizing processing, additional processings such as the fine line forming are usually necessary. Therefore, the processing system is complicated and enlarged in size, and it is difficult to realize speeding up.
The tracing method has the following problems. At the start of the tracing, since there is no clue to the tracing direction, the tracing is performed in a wrong direction in some cases. Since the object is traced without any target point, there is no guarantee that the tracing returns to a correct contour line at the time of tracing failure. Therefore, once the contour deviates, the tracing takes a winding course. When the tracing fails, the tracing is preferentially performed in the wrong direction, and continuously advances. Even when the correct contour line is returned, an error portion by the wrong tracing is emphasized.
The area limiting method has the following problems. Specifically, when the contour line is circular arc shaped, a plurality of short strip areas need to be set so that the contour line is uniformly included, which requires a very large number of processes. When the strip area can be curved, the operator is requested to trace along the contour line, or finely set reference points. Therefore, when the object has a complicated shape, a burden on the operator is large. In other words, the method is unsuitable for the object having a complicated shape.
Further, when the edge strength of the background running in the vicinity of the contour line included in the strip area is stronger than the edge strength of the true contour of the object, the background edge is wrongly detected. As a countermeasure, in the U.S. Pat. No. 5,617,487 is described a method which comprises the steps of setting a contour point determining weight distribution in the strip width direction of the strip area, and raising the edge detecting sensitivity in the vicinity of the center line of the strip area. In this method, the operator has to cautiously set the strip area so that the contour line is positioned in the vicinity of the center of the strip area. This is a complicated and laborious operation, and requires considerable skills. Of course, even when the weight distribution is set in this manner, in a case in which the edge strength of the background is strong, the background edge is wrongly detected. Therefore, this countermeasure is not perfect.
The active contour method has the following problems. Specifically, this method is sensitive to the precision of an initial value, that is, the initially set position of the closed curve. If the initial setting is bad, a correct contour is not converged. The calculation amount is usually enormous, and the speeding up is difficult. An energy function is defined on the assumption that the contour line is smooth, and the method is poor in extracting the contour line which changes in an acute angle.
SUMMARY OF THE INVENTION
An object of the present invention is to provide an image processing apparatus, method and storage medium to solve these problems.
Another object of the present invention is to provide an image processing apparatus, method and storage medium in which an image contour can highly precisely be extracted from a wide-range natural image including a complicated background with a simple operation at a high speed.
To achieve the above-described objects, a preferred embodiment of the present invention comprises a trace processing step of tracing a boundary of an image to be extracted among input images, and a limiting step of limiting a tracing
Kondo Toshiaki
Matsugu Masakazu
Takahashi Fumiaki
LandOfFree
Image processing apparatus, method and storage medium does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Image processing apparatus, method and storage medium, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Image processing apparatus, method and storage medium will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3155796