Image analysis – Pattern recognition – Feature extraction
Reexamination Certificate
1998-07-29
2001-10-02
Johns, Andrew W. (Department: 2723)
Image analysis
Pattern recognition
Feature extraction
Reexamination Certificate
active
06298155
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
This invention relates to an apparatus and a method for generating time-series data, an apparatus and a method for editing curves and a recording medium having a program recorded thereon.
2. Description of Related Art
The processing of extracting an object shape from plural picture data is frequently performed in computer-aided design (CAD), computer graphics and in a variety of picture processing operations. In particular, there is a processing of formulating a key signal required in effecting picture synthesis processing. This key signal, also termed a mask, is the information required in slicing an area of a foreground object desired to be synthesized. In picture synthesis, it is felt to be crucial to extract a particularly accurate key signal, that is accurate contour shape.
The relevant contour extracting technique may be classified in two as to the operating method.
The first relevant technique is to extract the contour automatically after initially affording the information for the contour. The information to be given first is the rough shape information of the contour by drawing a rough closed curve in the vicinity of the contour to be extracted or the information of the color or texture necessary for discriminating the contour.
The former may be exemplified by techniques such as “area extraction method” described in Japanese Laying-Open Patent H3-17680, “area extraction apparatus” described in Japanese Laying-Open Patent H5-61977 and “Contour Tracing method for moving objects” described in Japanese Laying-Open Patent H5-12443.
The latter may be exemplified by “slicing mask producing system” described in Japanese Laying-Open Patent H2-105152 and “slicing mask producing method and Apparatus” described in Japanese Laying-Open Patent H4-90544.
The second relevant technique is of the type in which an operator produces the shape as he or she sets the position or direction of the contour in detail and has more interactive operating characteristics. It is proposed to designate plural points on a contour to generate the shape between the points, to act on control points of a parametric curve representing the contour shape or to directly enter the contour shape by a mouse.
This technique is disclosed for example, in “picture contour detecting method” in Japanese Laying-Open Patent H4-152481 or in “slicing mask producing method and apparatus” in Japanese Laying-Open Patent H4-254854.
In the processing for generating a key or a mask, it is necessary that an accurate contour shape be obtained for a picture of the entire frame. In a motion picture or a television picture, since hundreds of keys are required for synthesizing a picture continuing for several seconds, the processing volume is significant. Therefore, contour extraction processing automatically executed with as small a number of inputs as possible, such as the first relevant technique, is desirable.
However, the current state of the art cannot achieve contour shape extraction completely, such that, for obtaining keys that can stand utilization to picture synthesis, it is indispensable that the operator partially corrects the contour by manual operation to realize a more accurate contour shape. As a technique for aiding this correction, an interactive technique such as the second relevant technique is desirable.
However, the following problem arises in the relevant contour extraction technique.
Basically, the relevant contour extraction technique effects independent processing from frame to frame to realize the contour shape for each frame. Thus, sufficient processing accuracy cannot be achieved depending on picture characteristics, such as luminance or bleeding, with the contour position changing delicately from frame to frame. Even if accurate contour extraction appears to be achieved, the contour line, which should exhibit smooth movement, tends to be fluctuated if the frame is viewed in its entirety.
Moreover, if contour extraction is done in each frame using an interactive technique as in the second relevant technique, operator input errors are also responsible for fluctuations. Since partial manual correction is indispensable in key generation for producing a key that can be used safely for picture generation, problems of frame-to-frame contour fluctuations is presented unavoidably.
In general, for suppressing fluctuations in time-series data, smoothing in the time axis is used frequently. This smoothing in the time axis is also effective in suppressing the above-mentioned frame-to-frame contour fluctuations. However, the method for smoothing contour shape data along the time axis suffers from the following inconveniences.
For effectuating smoothing, respective positions on the contour need to be associated from frame to frame. In the relevant contour extraction technique, frame-based processing is independent, as described above, such that frame-to-frame contour position correspondence is not realized, so that this correspondence needs to be realized separately.
The overall length of the contour differs from frame to frame, the shape is changed with time or data start position on the contour differs from frame to frame, depending on the particular contour data generating method employed, such that automatic corresponding point tracking becomes difficult.
If, for avoiding the above problem, a curve defined by a pre-set number of parameters is used, correspondence can be achieved from parameter to parameter, so that parameter-based smoothing is possible. However, since complexity of the shape of a curve that can be represented by a pre-set number of parameters is fixed, accurate contour extraction becomes infeasible.
On the other hand, it is extremely labor-consuming to give necessary corresponding points by a manual operation in their entirety.
It is seen from above that such a method is required which consists of giving a certain number of corresponding points manually and to estimate other corresponding points based on these given points.
Although the processing of smoothing time-series data is effective in suppressing data fluctuations, as described above, shape dullness results due to elimination of high frequency components of time-series data.
In the case of contour data in a picture, a phenomenon may arise in which contour data of a moving object cannot follow the contour of the object in the original picture to lead to significant deterioration of contour accuracy.
SUMMARY OF THE INVENTION
It is therefore an object of the present invention to provide a method and apparatus for generating time-series data, a curve editing method and apparatus and a recording medium having a program recorded thereon, in which, if a certain number of corresponding points are given manually, other corresponding points are estimated from the given points, and in which it is possible to evade the phenomenon where contour data of a moving object cannot follow the contour of the object in the original picture.
In one aspect, the present invention provides a time-series data generating apparatus for generating, from time series I(t) of data distributed in space, time series P(t) of corresponding coordinates, including interpolation curve generating means for receiving an incomplete set of the corresponding coordinates as input for generating an interpolation curve C(t) interconnecting the coordinates, candidate point extraction means for receiving the interpolation curve C(t) and a time series I(t) of data as inputs for extracting a group of candidate points Q of a coordinate P(ts) to be found at a time juncture ts devoid of a corresponding coordinate from the neighborhood of a position C(ts) of the interpolation curve at time ts, and evaluated value calculating means for receiving the time ts, interpolation curve C(t), time series of data I(t) and a candidate point q of the group of candidate points as inputs for calculating an evaluated value e(q) for the candidate point q, and candidate point selection means for receiving the evaluation value e(q) as an input for detecting th
Hirase Hidehiro
Mitsunaga Tomoo
Totsuka Takashi
Yokoyama Taku
Zeiss John
Azarian Seyed
Bell Boyd & Lloyd LLC
Johns Andrew W.
Sony Corporation
LandOfFree
Apparatus and method for generating time-series data,... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatus and method for generating time-series data,..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatus and method for generating time-series data,... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2609905