JPH07107370A - Motion vector detector - Google Patents

Abstract

PURPOSE:To obtain a motion vector of high precision by correctly obtaining the slop of even a picture signal with many high frequency components such as edge components in a motion vector detector extracting motion vector information from plural input picture signals. CONSTITUTION:A low pass filter 6 removes the high frequency components included in the input signal. Then, a slope calculation part 3 obtains the slope of a picture from the input signal obtained by removing the high frequency components. The low pass filter 6 is provided with a filter characteristic corresponding to the slope calculation of the picture in the slope calculation part 3.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion vector detecting device for extracting motion vector information from a plurality of input image signals, and more particularly to a motion vector detecting device suitable for use in a camera shake correcting device of a camera device.

[0002]

2. Description of the Related Art The above-described conventional motion vector detecting device obtains a motion vector by using a matching algorithm such as representative point matching or block matching, or a gradient method. The gradient method will be described below. t
= 0, an image g (x, y) moves to t = τ by a vector α _v = (a, b) and g (x−a, y−).
b). Here, it is assumed that α _v is very small, and the Taylor expansion of g (x−a, y−b) is performed, and only the first-order term is set.

[0003]

[Equation 1]

Equation (2) is derived from this equation (1).

[0005]

[Equation 2]

Therefore, the motion vector α _v can be obtained by solving the simultaneous equations by establishing the equation (2) for two different images in the image g (x, y). By the way, since the accuracy is remarkably deteriorated if it is calculated from only two points, the images g (x, y) and g (x-a, y-b) are divided into blocks of an appropriate size, and each block is dealt with. Equation (2) is established for each point in the block, and the motion vector α _v is obtained using the least square approximation.

FIG. 7 is a block diagram showing a schematic configuration of a conventional motion vector detecting device using the gradient method. In this figure, 1 _i (i = 1,2,3, ..., n) is one block constituting the input image 1 and 2 _i (i = 1,2,3, ..., n) is the input image 2 One of the blocks constituting the unit, 3 is a gradient calculating unit for calculating the gradient in the formula (2), 4 is an image difference calculating unit for calculating the image difference Δd (x, y), and 5 is a least-squares approximation. It is a motion vector calculation unit that calculates the motion vector α _v .
Each block 1 _i of the input image 1 is supplied to the gradient calculation unit 3. Each block 1 of the input image 1 is stored in the image difference calculation unit 4.
_i and each block 2 _i of the input image 2 are supplied. The motion vector calculation unit 5 is supplied with the gradient calculated by the gradient calculation unit 3 (see FIG. 8) and the inter-image difference Δd (x, y) calculated by the inter-image difference calculation unit 4. FIG. 9 is a diagram showing a distribution of motion vectors obtained by a conventional motion vector detection device.

[0008]

By the way, in the above-mentioned conventional motion vector detecting device, since the gradient is approximately obtained from the value of the pixel adjacent to the point of interest, the edge component etc. In an image with many high frequency components, an error occurs in the gradient calculation, and there is a problem that a highly accurate motion vector cannot be obtained. To be more specific, it is assumed that n is an appropriate integer and the image gradient at a certain point (x, y) is calculated by the following equation.

[0009]

[Equation 3]

Here, the sampling frequency of the image is set to f _s
Then, when n = 1 in the equation (3), a signal having a frequency of f _s / 4 or more cannot be expressed in the pixel adjacent to the point (x, y) of interest (see FIG. 10). And in this case,
Since the Nyquist frequency is f _s / 2 and a signal whose frequency is f _s / 4 or more is included, aliasing occurs and an error occurs in the gradient calculation. Similarly, when n = k, a signal of f _s / (4 × k) or more cannot be expressed, and thus an error also occurs at this time. Due to these errors, a highly accurate motion vector cannot be obtained.

Therefore, according to the present invention, the gradient can be accurately obtained even for an image signal having many high frequency components such as edge components.
It is an object of the present invention to provide a motion vector detecting device which can obtain a highly accurate motion vector.

[0012]

To achieve the above object, a motion vector detecting device according to the present invention is a motion vector detecting device for extracting motion vector information from a plurality of input image signals, and is included in the input image signals. It is characterized by comprising a filter section for removing high frequency components, and a gradient calculation section for obtaining an image gradient using an output signal of the filter section. In a preferred aspect, the filter unit has a filter characteristic corresponding to the gradient calculation of the image in the gradient calculation unit.

[0013]

In the present invention, the input image signal is band-limited by the filter to remove high frequency components such as edge components. After that, the motion vector is detected by the gradient method. Therefore, even with an image signal having many high-frequency components such as edge components, the gradient can be accurately obtained, and a highly accurate motion vector can be obtained.

[0014]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the drawings. Example 1. 1 is a block diagram showing a schematic configuration of a first embodiment of a motion vector detection device of the present invention. The same parts as those in FIG. 7 described above are designated by the same reference numerals and the description thereof will be omitted. In this figure, reference numeral 6 denotes a one-dimensional low-pass filter (LPF) having a filter characteristic corresponding to the gradient calculation of the image in the gradient calculation unit 3, which filters in the x and y directions independently. Each block 1 _i (i = 1,2,3,
..., n) and each block 2 _i (i = 1,2,3, ...,) of the input image 2
n) and are band-limited horizontally and vertically. The cutoff frequency of the low-pass filter 6 is f _s when n = k.
/ (4 × k). By limiting the band of the input image with the low-pass filter 6, edge components and the like of the image are removed.

Each block 1 _i ′ of the input image band-limited by the low-pass filter 6 is supplied to the gradient calculating section 3 and the inter-image difference calculating section 4. At the same time, each block 2 _i ′ of the input image whose band is limited by the low-pass filter 6 is supplied to the inter-image difference calculation unit 4. Then, the gradient calculated by the gradient calculation unit 3 and the inter-image difference Δd (x, y) calculated by the inter-image difference calculation unit 4 are supplied to the motion vector calculation unit 5, and the least-squares approximation is performed. By the motion vector α
_v is calculated. Here, FIG. 2 is a diagram showing the gradient calculation when the band is limited.

Example 2. FIG. 3 is a block diagram showing a schematic configuration of a second embodiment of the motion vector detection device 110 of the present invention. The second embodiment applies the motion vector detecting device 100 of the first embodiment, and the motion vector detecting device 1
, N is obtained on the basis of the motion vectors 1, 2, 3 ,. In describing the motion vector detection device 110, it is assumed that there are two input images and the entire images are shifted by the same amount in the same direction. Further, in the motion vector detection by the gradient method, when the motion vector becomes large, the approximation of the equation (1) cannot be established and an error occurs, so the shift amount is set to a value smaller than one pixel.

In the gradient calculator 3 (see FIG. 1), the gradient of the image is approximated by the equation (3) with n = 1. Further, the cutoff frequency of the low pass filter 6 is set to f _s / 4
I have to. After dividing the input image into an appropriate number of blocks, the motion vector 1, 2,
3, ..., N are detected. Then, the shift amount is obtained based on the obtained motion vectors 1, 2, 3, ..., N. At this time, the shift amount is obtained by a method of weighting with a square error and taking an average. In addition to this method, the following 2
There is one. Take a simple average. An average is obtained by weighting with a squared error in the least-squares approximation.

Here, in order to show the effectiveness of the motion vector detecting device 110, two images having a shift amount of one pixel or less were generated and the shift amount was detected. Input image is 3
It was created by thinning out 072 × 2048 images to 1/8.
The image can be shifted in 1/8 pixel increments by changing the phase during thinning. Here, the input image is shifted by 2/8 pixels both vertically and horizontally. FIG. 4 is a motion vector distribution diagram obtained by the motion vector detection device 110, and it can be seen that the motion vector is not dispersed as compared with the conventional motion vector distribution diagram shown in FIG. Also, FIG.
6 is a diagram showing a shift amount obtained by the motion vector detection device 110 and a shift amount obtained by a conventional motion vector detection device. As can be seen from this figure, the motion vector detection device 110 is more accurate. Good vector detection.

Example 3. FIG. 6 is a block diagram showing a schematic configuration of a third embodiment of the motion vector detection device 120 of the present invention. The third embodiment is provided with a gradient calculating section 7 using an FIR filter. The same parts as those of the first embodiment described above are designated by the same reference numerals and the description thereof will be omitted. The first embodiment can also be realized by an FIR filter, which is shown in the equation (3) when n = 1.
In the gradient calculating unit 3 of the first embodiment, the calculation of the gradient by the equation (3) is performed by the tap coefficient F shown in the equation (4).
It is equivalent to operating an IR filter.

[0020]

[Equation 4]

Further, in order to perform band limitation f _3/4, using a FIR filter tap coefficients shown in equation (5) as a low-pass filter.

[0022]

[Equation 5]

Therefore, it is finally equivalent to applying the FIR filter having the tap coefficient shown in the equation (6) which is the convolution of the tap coefficients of the two FIR filters.

[0024]

[Equation 6]

As a result, the gradient may be calculated according to the equation (7).

[0026]

[Equation 7]

As described above, when the FIR filter is used, the gradient can be calculated directly according to the tap coefficient obtained by convolving the tap coefficient of the low-pass filter and the tap coefficient of the gradient calculation.

Although the second embodiment is the one to which the first embodiment is applied, the third embodiment may be applied.

[0029]

According to the present invention, the input image signal is band-limited by the filter to remove the edge component and the like of the image, and thereafter the motion vector is detected by the gradient method. Even if the image signal has many high frequency components, the gradient can be accurately obtained, and a highly accurate motion vector can be obtained.

[Brief description of drawings]

FIG. 1 is a first embodiment of a motion vector detection device according to the present invention.
It is a block diagram of.

FIG. 2 is a diagram showing gradient calculation of the motion vector detection device according to the first embodiment.

FIG. 3 is a second embodiment of the motion vector detecting device according to the present invention.
It is a block diagram of.

FIG. 4 is a distribution diagram of motion vectors obtained by the motion vector detection device according to the second embodiment.

FIG. 5 is a diagram showing a difference between the motion vector detecting device according to the second embodiment and a conventional motion vector detecting device.

FIG. 6 is a third embodiment of the motion vector detecting device according to the present invention.
It is a block diagram of.

FIG. 7 is a block diagram of a conventional motion vector detection device.

FIG. 8 is a diagram showing a gradient of a vector of a conventional motion vector detecting device.

FIG. 9 is a distribution diagram of motion vectors obtained by a conventional motion vector detection device.

FIG. 10 is a diagram showing gradient calculation of a conventional motion vector detection device.

[Explanation of symbols]

 6 Low-pass filter (filter section) 3, 7 Gradient calculation section

Claims (2)

[Claims] 1. A motion vector detection device for extracting motion vector information from a plurality of input image signals, comprising: a filter unit for removing high frequency components contained in the input image signal; and an output signal of the filter unit. A motion vector detection device comprising: a gradient calculation unit that obtains an image gradient. 2. The motion vector detection device according to claim 1, wherein the filter section has a filter characteristic corresponding to the gradient calculation of the image in the gradient calculation section.