Image analysis – Color image processing – Pattern recognition or classification using color
Reexamination Certificate
2001-09-19
2003-06-24
Patel, Jayanti K. (Department: 2625)
Image analysis
Color image processing
Pattern recognition or classification using color
C358S538000, C375S296000
Reexamination Certificate
active
06584222
ABSTRACT:
BACKGROUND OF THE INVENTION
The present invention relates to a feature-region extraction method and a feature-region extraction circuit and, more particularly, to a feature-region extraction method and a feature-region extraction circuit which are capable of extracting a feature-region of an image for processing the image according to its feature at a pre-processing portion of an image coding device and which are intended to be used, for example, in video telephone systems, video conference systems and so on.
At a motion picture coding and transmitting device for a video telephone, since any transmission line has a limited or finite capacity to transmit a number of bits at a time, it cannot allocate enough bits to an image information. Accordingly, a decoded image may have an impaired picture quality with a mosquito-like distortion, a block distortion and the like. On the other hand, most of the images to be transmitted include an important portion and less important portions. On the basis of this fact, a method is proposed that reduces the subjective deterioration of an image quality by keeping the important portion free from the ill effects of deterioration.
For example, in systems such as video telephone systems, the image of the face region is so important that such systems have been directed toward extracting the face region, preferentially allocating the amount of information transmitted and making the image quality of the face region better, aiming at improvement of the subjective image quality.
The prior art that is concerned with the present invention is disclosed in the publication “Color motion picture coding device featured by preferentially processing a face image” (Hideyuki Ueno, Image Information vol. 24, March 1992, pp. 29-35). The face-region extraction method of the prior art described in the publication will be explained as follows.
The prior art first determines the difference between two neighboring frames of the motion picture inputted to get the interframe differential image and then divides the interframe differential image into rectangular regions. In the respective rectangular regions, the method makes a histogram representing the number of pixels, in which the interframe difference signal is larger than a certain threshold, on an H-axis. The method scans the histogram from the left to right, comparing its values with a certain threshold “th” to detect the left and right ends of the moving region.
The image screen is scanned from the upper left end until a first moving region is found. This moving region has a coordinate on a V-axis, which represents a position at the top of one's head. The image screen is then scanned from the left and the right by a certain range of lower image to determine the left and right ends of the moving region. And then, the outermost positions of the moving region on the H-axis represent the coordinates of the left and right ends of the face. The face length is determined by multiplying the face width by a predetermined coefficient “&agr;”. The prior art considers the thus determined image region as a face-region.
As mentioned above, the conventional feature-region extracting method estimates a face-region from an interframe difference signal representing the amount of motion. Consequently, if the image includes any other moving object than the face-region, the prior art method may erroneously take the object as a face part. Furthermore, the method can roughly extract a face-region only in the form of a rectangle, which contour may be emphasized when the regions other than the face are adaptively processed with a low-pass filter. This may impair the subjective image quality.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide a feature-region extraction method and the feature-region extraction circuit, which can extract a feature-region exactly along its contour without erroneously extracting any other moving object than the feature one, thereby allowing adaptive processing of portions other than the feature-region without impairing the image quality and can extract a face-region, using color difference components, regardless of the movement and location of a person or a number of persons, improving the subjective image quality.
It is another object of the present invention to provide a feature-region extraction method whereby a small color difference region including a feature color difference component of a feature-region is located within a color difference coordinate system defined by two axes representing two respective color difference signals and an image region, which has a color difference component in the small color difference region mentioned above, is extracted as a feature-region; the small color difference region can take any desired form thereby eliminating the possibility of erroneously extracting any moving object other than the feature-region; the feature-region can be exactly extracted along its contour line to make the image more natural and to have a subjectively improved quality and, therefore, can withstand adaptive processing of other regions with a low-pass filter; and, furthermore, only color difference signals are applied for feature-region extraction to prevent erroneous extraction when the luminance level of the feature-region varies in space and in time.
It is another object of the present invention to provide a feature-region extraction method whereby a small color difference region is a common region of two color difference signals binarized by the use of specified thresholds and a feature-region is easily extracted therefrom.
It is another object of the present invention to provide a feature-region extraction method whereby a feature-region is extracted from a plurality of small color difference regions so as to further improve the image quality.
It is another object of the present invention to provide a feature-region extraction method whereby a feature-region along its contour line is extracted together with a dropout region enclosed within the feature-region contour line, so as to further improve the image quality.
It is another object of the present invention to provide a feature-region extraction method whereby a feature-region is extracted from a common region of two color difference signals binarized by the use of specified thresholds and dropout regions within the feature-region are filled up by using detection patterns weighted by distance so as to further improve the image quality.
It is another object of the present invention to provide a feature-region extraction method whereby in extracting a feature-region by using means for extracting a feature-region and a memory means for storing a feature-region extraction signal, a criterion is given a hysteresis characteristic depending upon whether a pixel backed by L frames (L is a positive integer) and/or its neighbor pixel is a feature-region or not, making it possible to effectively prevent the feature-region from being extracted with flickers and thereby to further improve the subjective image quality.
It is another object of the present invention to provide a feature-region extraction method whereby a feature-region is extracted by using its characteristic, and feature-region extraction signals of plural frames from a memory of M frames (M is a positive integer) are weighted for extracting a weighted feature-region so as to further improve the image quality.
It is another object of the present invention to provide a feature-region extraction circuit whereby a feature-region extraction circuit that includes a means for extracting a main portion of a feature-region and for filling a dropout region can extract a feature-region exactly along its contour line, and thereby adaptive processing of an image region other than the featured one with a low-pass filter can be performed free from the image deterioration.
A feature-region extracting method, according to the present invention, can extract a feature-region by using a small color difference region including a feature color difference component of a fea
Conlin, Esq. David G.
Edwards & Angell LLP
Patel Jayanti K.
Ross, Esq. Richard J.
Sharp Kabushiki Kaisha
LandOfFree
Feature-region extraction method and feature-region... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Feature-region extraction method and feature-region..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Feature-region extraction method and feature-region... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3139349