Image analysis – Applications – Motion or velocity measuring
Reexamination Certificate
1998-10-01
2001-07-17
Johns, Andrew W. (Department: 2621)
Image analysis
Applications
Motion or velocity measuring
Reexamination Certificate
active
06263089
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention generally relates to techniques for recognizing a target within an image sequence, and more particularly to a method and an equipment for extracting image features from the image sequence which describes a time sequence of frames of the image.
The image sequence refers to an image which is obtained from a video camera, weather radar equipment, remote sensing or the like, for the purposes of monitoring people, traffic and the like, controlling fabrication processes, analyzing or predicting natural phenomena such as the weather.
2. Background Art
Local (for example, several tens to several hundreds of km
2
) and short-term (for example, 5 minutes to several hours) precipitation phenomena such as heavy rain, heavy snow and thunderstorm have yet to be elucidated completely. However, the effects of the local and short-term precipitation phenomena on daily lives and various industrial activities are large, and it is an important task to predict the precipitation phenomena.
Conventionally, in order to forecast such local precipitation phenomena, an expert such as a meteorologist visually specifies the phenomena from an observed weather radar image and creates a weather forecast. In addition, the weather forecast is created by analyzing a motion of an echo pattern within a weather radar image, and referring to a predicted echo image which is obtained by predicting a future echo pattern. The former prediction is based on the regularity of the weather phenomena acquired by the expert from past experiences, and requires years of skill. On the other hand, according to the latter prediction using image analysis, it is assumed in most cases that the phenomenon of immediately preceding several hours is maintained, and it is thus impossible to follow a rapid change in the phenomenon even though the forecast most expected to predict such a rapid change. Furthermore, because it is impossible to satisfactorily represent the phenomena such as an accurate moving velocity, appearance, disappearance, deformation and the like of a precipitation region, there is a problem in that the prediction accuracy is insufficient.
Accordingly, as one method of making an improvement with respect to the above described problem, it is conceivable to utilize a repeatability of the weather phenomena that “similar weather phenomena occur repeatedly”, and to automatically retrieve past weather radar images with similar phenomenons based on the weather radar image, so as to present the similar past weather radar images to the expert. Alternatively, it is conceivable to categorize the weather radar images into categories of the weather phenomena, and to select and apply a prediction technique suited for each specified weather phenomenon. In order to realize such methods, it is necessary to extract an image feature value (hereinafter also simply referred to as an image feature) from the weather radar image which is an image sequence data.
Conventionally, as methods of extracting the image feature of the image sequence, texture analysis techniques which obtain the features of a texture within a still image, and motion estimation techniques which obtain a displacement quantity of the image pattern between frames of the image sequence have been proposed.
For example, Robert M. Haralick, “Statistical and Structural Approaches to Texture”, Proceedings of the IEEE, Vol.67, No.5, May 1979 proposes a statistical texture analysis which is one approach of the conventional texture analysis technique. According to this statistical texture analysis, statistics such as “a frequency of existence of a combination of a certain pixel and another pixel located 3 pixels to the right of the certain pixel having a luminance difference of 1 between the certain pixel and the other pixel” is calculated, and the image features are extracted. This statistical texture analysis is used to detect a difference in two-dimensional image features such as a pattern (called “texture”) on the image surface obtained by a repetition of basic graphic elements. More particularly, a set of basic elements called primitives is first obtained from the image of 1 frame of the image sequence by a process such as image binarization. Next, a spatial feature such as directionality is calculated as the statistics such as the direction and length of an edge of each primitive. In addition, the spatial feature such as the regularity of the above described repetition of the primitives is calculated from relative position vectors among the primitives.
The image feature proposed by Robert M. Haralick referred above includes a feature value which is defined from a co-occurrence matrix of the image gray level. The co-occurrence matrix is a matrix having as its element a probability P
&dgr;
(i, j), (i, j=0, 1, . . . , n−1) that a point which is separated by a constant displacement &dgr;=(r, &thgr;) from a point having a gray level (or brightness or intensity) i in the image has a gray level j. For example, feature values such as those described by the following formulas (0.1) and (0.2) can be calculated from the co-occurrence matrix, where &dgr; is set to r=1, &thgr;=0 (deg), for example.
angular
second
moment
=
∑
i
=
0
n
-
1
∑
j
=
0
n
-
1
{
P
δ
(
i
,
j
)
}
2
(
0.1
)
entropy
=
-
∑
i
=
0
n
-
1
∑
j
=
0
n
-
1
P
δ
(
i
,
j
)
·
log
{
P
δ
(
i
,
j
)
}
(
0.2
)
The angular second moment described by the formula (0.1) represents the concentration and distribution of the elements of the co-occurrence matrix, and it is possible to measure the uniformity of the texture. Such a feature value is used to analyze the geographical features from an air photograph and sandstone. However, in general, the feature value obtained from the co-occurrence matrix is in many cases unclear as to what is being physically measured.
According to the conventional technique using the texture analysis, each frame of the image sequence is treated as an independent image. For this reason, no measurement is made with respect to the features related to the motion, although the motion is an essential element in determining the features of the image sequence.
On the other hand, as conventional motion estimation methods, Yoshio Asuma et al., “A Method for Estimating the Advection Velocity of Radar Echoes Using a Simple Weather Radar System”, Geophysical Bulletin of Hokkaido University, Sapporo, Japan, Vol.44, October 1984, pp.23-34 or Yoshio Asuma et al., “Short-Term Prediction Experiment (Part 1) of Snow Precipitation Using a Simple Weather Radar System”, Geophysical Bulletin of Hokkaido University, Sapporo, Japan, Vol.44, October 1984, pp.35-51 propose methods of obtaining 2 frames of the image sequence, matching each small region within the frames, and measuring the motion (velocity component) of a target included in the small region, for example. These proposed methods use the images of 2 different frames of the image sequence. First, a best matching position where a certain region (normally, a square region) within the image of one frame best matches the image of the other frame is searched. Next, the moving velocity of the object within the target region is estimated from a displacement between the 2 frames and the frame interval of the 2 frames. A cross-correlation coefficient of the image gray level value is used to describe the degree of matching of the 2 image regions. When the gray level distributions within the 2 image regions are respectively denoted by I
1
(i, j) and I
2
(i, j), the cross-correlation coefficient can be: calculated from the following formulas (0.3), (0.4) and (0.5), where M and N indicate the sizes of the 2 image regions.
σ
=
[
∑
i
=
1
M
∑
j
=
1
N
(
I
1
(
i
,
j
)
I
2
(
i
,
j
)
-
MN
I
_
1
I
_
2
]
/
[
(
∑
i
=
1
M
∑
j
=
1
N
I
1
(
i
,
j
)
2
-
MN
I
_
Horikoshi Tsutomu
Otsuka Kazuhiro
Suzuki Satoshi
Johns Andrew W.
Kenyon & Kenyon
Nakhjavan Shervin
Nippon Telephone and Telegraph Corporation
LandOfFree
Method and equipment for extracting image features from... 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 equipment for extracting image features from..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and equipment for extracting image features from... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2449204