KR100930883B1 - How to calculate meat fat content from ultrasound image - Google Patents

Abstract

The present invention relates to a method for calculating the fat content of meat from an ultrasound image, and extracts image features of textures such as edges and valleys from ultrasound images so as not to be sensitive to changes in brightness caused by the reflection and shadowing effects of ultrasound. By calculating the fat content of meat in consideration of the actual condition of meat quality according to the actual situation, the meat content is estimated without calculating slaughter, and the fat content of meat is estimated to estimate the characteristics of livestock. It has a predictable and adjustable effect.

Ultrasound, Imaging, ROI, Region of Interest, BDIP, BVLC

Description

METHODS MEASURING FAT CONTENT IN MEAT FROM ULTRASOUND VISUAL IMAGE}

The present invention relates to a method for calculating meat fat content from an ultrasound image, and more particularly, to extract image characteristics of textures such as edges and valleys from an ultrasound image so as not to be sensitive to a change in brightness caused by the reflection and shading effects of the ultrasound. Therefore, the present invention relates to a method for calculating the fat content of meat from an ultrasound image, which can estimate and calculate the fat content of meat in consideration of the actual state of meat according to the characteristics of livestock.

Ultrasonic imaging equipment, unlike other imaging equipments, does not harm the living body, and it is easy to continuously observe the inside of the living body for a long time in real time. In addition, it has a low price and small size compared to other equipments, and is widely used for diagnosing and treating diseases of animals and livestock.

Many attempts have been made to determine the fat content of livestock and to preserve and release excellent seed by utilizing the characteristics of these ultrasound imaging equipment.

To date, most domestic experts have determined fat content on ultrasound images qualitatively or by using fat content determination programs of foreign ultrasonic equipment companies such as Esaote Piemedical.

In addition, Iowa State University has established a region of interest (ROI) of 100 pixels in width and height at a location where image characteristics are well represented on an ultrasound image, and then uses various image processing techniques for the region of interest. Extract the features and estimate the fat content using the regression equation.

The image processing technique used to extract the image features used here is Fourier transform, gradient, histogram, co-occurrence matrix.

However, these image processing techniques are the most traditional methods of extracting image features such as edges and valleys from general images. They are somewhat difficult to use for special ultrasound images that are noisy and partially shaded like ultrasonic images. There is an unsuitable problem.

In addition, if you simply set the box size of 100 pixels in the ROI setting, the total size and distribution of muscles that can be set by the expert can not be used to determine fat content. There is a problem that the fat content value which is estimated to be very sensitive varies depending on the position of.

The present invention was created to solve the above problems, and an object of the present invention is to extract image features for textures such as edges and valleys from ultrasonic images so as not to be sensitive to brightness variations caused by the reflection and shadowing effects of ultrasonic waves. Therefore, the present invention provides a method for calculating the fat content of meat so that the fat content of meat can be estimated and calculated to reflect the actual condition of meat according to the characteristics of livestock.

The present invention for achieving the above object is the step of receiving and obtaining the characteristics of the livestock with the meat ultrasound image of the livestock, and after setting the primary region of interest in an arbitrary form surrounding the muscles in the ultrasound image of the primary region of interest Setting a secondary ROI for expressing a feature of an image based on the PDU, obtaining a normalized image by extracting the secondary ROI, performing normalization on the brightness of the image, and obtaining a normalized image within the block. Image features are analyzed by the BDIP technique, which normalizes the difference between the maximum brightness value and the brightness value of each pixel to the maximum brightness value, and the BVLC technique, which measures the change in the position correlation coefficient along four directions adjacent to the block. The fat content is calculated by applying the image feature enhancement step and the image feature values extracted by image feature enhancement to the regression equation according to the characteristics of the livestock. Characterized in that it comprises the step of shipping.

In the present invention, the characteristics of the livestock are characterized by being castrated, non-steered, and cow.

In the present invention, the secondary region of interest is set to a rectangle circumscribed to the primary region of interest, or set to a square circumscribed to the primary region of interest, or after calculating the center of gravity of the primary region of interest, a predetermined size based on this center point. It is characterized by setting to the square of.

delete

delete

As described above, the present invention extracts image characteristics of textures such as edges and valleys from ultrasonic images to reflect the actual state of meat according to the characteristics of livestock, so as not to be sensitive to changes in brightness caused by the reflection and shading effects of ultrasonic waves. By estimating the fat content of meat to be suitable for the practical situation, it is possible to calculate the fat content of meat without estimating livestock and to estimate the characteristics of livestock, thereby predicting and controlling the delivery time required for high-quality meat production.

In addition, the present invention can accurately estimate the breeding characteristics in a biological state has the effect of promoting breeding by forming a good breeding population and selecting a good breeder.

Hereinafter, exemplary embodiments of the present invention will be described with reference to the accompanying drawings, and the same parts as in the prior art use the same reference numerals and names. In addition, the present embodiment is not intended to limit the scope of the present invention, but is presented by way of example only and those skilled in the art will be capable of many modifications within the technical spirit of the present invention.

1 is a flowchart illustrating a method for calculating fat content of meat according to the present invention.

The present invention is to receive the characteristics of the Korean beef along with the ultrasound image taken between the thirteenth rib and the first lumbar spine to measure the intramuscular fat of the Korean beef (S10) (S12).

The characteristics of Hanwoo are castrated, non-stocked, cow, etc., so that it is possible to accurately estimate the characteristics of the distribution of intramuscular fat according to each characteristic.

The primary region of interest is set in an arbitrary form as shown in FIG. 2 for the region surrounding the muscle where the intramuscular fat is distributed in consideration of the change in brightness of the leather or the boundary of the input ultrasound image (S14). ).

The primary region of interest may be determined by an expert to determine an area for calculating intramuscular fat through images.

When the primary ROI is set, the brightness of the image is generated at each position even within the primary ROI. To compensate for this, as shown in FIG. Will be set.
This is because the distribution of fat for each location is very different even if it is set by an ultrasonic meat quality diagnosis expert for the primary area of interest, and the fat content value varies greatly from location to location. For this reason, as in the conventional method, if the rectangular ROI of a certain size (100 pixels x 100 pixels) is placed at an arbitrary position in the primary region of interest, the fat content is largely out of reality.
Therefore, it is essential to properly set a standardized ROI based on the entire primary area of interest in order to more accurately evaluate the total intramuscular fat. In addition, since the primary region of interest is set by the naked eye in an ultrasound image having a low resolution, a slight difference may occur even if the same image is repeatedly set. In order to minimize these changes and reduce the complexity of the calculation, a blue ROI having a different size, that is, a secondary ROI, is set based on the primary ROI shown in red as shown in FIG. 3.

In this case, the secondary ROI may be set to a quadrangle circumscribed to the primary ROI as shown in FIG. 3A, and may be set to a quadrangle inscribed to the primary ROI as shown in (b). .

In addition, as shown in (c), it may be set by moving within a predetermined interval from the primary region of interest, and as shown in (d), the center of gravity of the primary region of interest is calculated and then fixed based on the center point. It can also be set to a rectangle of size.
Preferably, the secondary ROI may be set by the method of FIGS. 3A, 3B, and 3D to reduce the complexity of the calculation.

After setting the ROIs in the 1st and 2nd phases, the image of the ROI is extracted and normalized to the brightness of the image.

That is, since the brightness value or the standard deviation of the image may vary depending on the environment, the average brightness value and the standard deviation of the image are normalized by using Equation 1 to eliminate the difference. In operation S16, a normalized image is obtained.

I (x, y): Brightness value at pixel (x, y) in original image

μ: Average brightness value of original image

σ: standard deviation of the brightness of the original image

: 정규화된 영상 내의 화소(x,y)에서의 밝기값

: Brightness value at pixel (x, y) in normalized image

: 정규화된 영상의 평균 밝기값

: Average brightness value of normalized image

: 정규화된 영상의 밝기의 표준편차

: Standard deviation of brightness of normalized image

As described above, the image characteristic is emphasized by the image processing technique for each block on the normalized image having normalized the brightness of the image (S18).

That is, the shape or texture of an object in the image is extracted as an edge or a valley, and it is necessary to emphasize the feature of the image so that the feature can be extracted better.

Common image processing techniques use edge operators such as Fourier transform, discrete cosine transform (DCT), sobel or gradient, which are not only sensitive to changes in brightness, but also when using edge operators to extract edges and valleys. Since the characteristics vary depending on the size of the extracted value, a significant difference occurs when the size of the edge band and valley is difficult to extract the edge and valley at the same time.

Accordingly, in the present invention, a block difference of inverse probabilities (BDIP) technique for normalizing the average of the difference between the maximum brightness value in each block and each pixel brightness value to the maximum brightness value, and a change in the position correlation coefficient along four directions adjacent to the block By using the block variation of local correlation coefficients (BVLC) technique, an image with emphasis on image features that can easily extract edges and valleys can be obtained.

Hereinafter, the BDIP technique and the BVLC technique will be described.

First, the BDIP method sets up a square local window of size W at an arbitrary position (x, y) in the image and calculates the difference of each pixel brightness value with respect to the maximum brightness value IM (x, y) of the local area. It is defined as the average of the calculations and the value divided by the maximum brightness of the local area.

BDIP (x, y): BDIP value in pixel (x, y)

I (x, y): Brightness value in pixel (x, y)

W: Peripheral area centered on pixel (x, y)

: 영역 W 내의 화소의 개수

: The number of pixels in the area W

I _M (x, y): Maximum brightness in the area centered on pixel (x, y)

As such, the BDIP method calculates the difference of the brightness value of each pixel with respect to the maximum brightness value of the local area in the molecule, so that the edge or valley value can be extracted regardless of the direction and the maximum brightness value of the local area in the denominator. By normalizing, the edges and valleys in the dark region are more sensitively extracted, which makes it easier to extract the features of an ultrasound image that is noisy and has a high brightness change even in the image.

In addition, the BVLC technique first uses local correlation coefficients between a square window of size W at any position (x, y) and a window spaced apart by (k, l) as shown in Equation 3. )

Here, the value of (k, l) is set in four directions as shown in FIG. 4, and after each correlation coefficient is obtained, the value obtained by subtracting the minimum value from the maximum value is defined as BVLC, as shown in Equation 4.

BVLC (x, y): BVLC value in pixel (x, y)

ρ (x, y): correlation coefficient

μ, σ: mean and standard deviation

: 영역 W 내의 화소의 개수

: The number of pixels in the area W

As described above, the BVLC method emphasizes the difference in the correlation coefficient between adjacent regions so that the roughness of the texture in the image can be extracted more effectively.

If BDIP and BVLC are applied to the region of interest, an image with emphasis on image features such as edges, valleys, and textures can be obtained.Average and standard deviation of the image are applied to a regression equation such as Equation 6 to express fat content. It is calculated (S20).

Where μ and sigma represent the mean and standard deviation, respectively, and K represents the window with a size from 1 to 8.

In this case, the regression equation is defined according to the characteristics of the livestock, and a plurality of regression equations defined by constants reflecting the characteristic values are defined and applied according to the characteristics of the selected livestock.

Figure 5 is a graph comparing the fat content according to the domestic fat of the livestock with the actual fat content of the livestock (a) is the fat content calculated using the edge operator for 60 or more castrated Hanwoo cattle, (b) is the present invention It is a graph showing the fat content calculated by the method.

In the graph, the x-axis represents the actual measured fat content and the y-axis represents the estimated fat content.

As shown in the graph, the closer the result graph between the estimated fat content and the measured fat content is to the diagonal, the more accurately the graph is estimated by the method according to the present invention. When the correlation was examined, the correlation was 0.83, and the correlation by the conventional method using the edge operator was about 0.72.

1 is a flow chart illustrating a method for calculating the fat content of meat from an ultrasound image according to the present invention.

2 is a meat ultrasound image of livestock according to the present invention.

3 is an ultrasound image photograph of a region of interest set according to the method of the present invention.

4 is a diagram illustrating the movement of a window to obtain a four-way correlation coefficient for a 2x2 window by the BVLC method according to the present invention.

5 is a graph comparing the fat content calculated by the present invention.

Claims (5)

Obtaining and acquiring the characteristics of the livestock along with the ultrasound image of the livestock meat; Setting a primary ROI in an arbitrary shape surrounding the muscles in the ultrasound image, and then setting a secondary ROI for expressing features of the image based on the primary ROI; Extracting the second region of interest and normalizing the brightness of the image to obtain a normalized image; BDIP technique which normalizes the average of the difference between the maximum brightness value in each block and the brightness value of each pixel with respect to the normalized image to the maximum brightness value, and a block of BVLC technique that measures a change in the position correlation coefficient along four directions adjacent to the block. An image feature emphasis step of emphasizing the image feature by a star image processing technique; Calculating fat content by applying the image feature values extracted by the image feature emphasis to a regression equation according to the characteristics of the livestock Method for calculating the fat content of meat from an ultrasound image, characterized in that consisting of. The method of claim 1, wherein the characteristics of the livestock are castration, non-gerration and cow. delete The method of claim 1, wherein the secondary region of interest is A quadrangle that is circumscribed to the primary region of interest, or a quadrangle that is inscribed to the primary region of interest, or after calculating the center of gravity of the primary region of interest, is set to a square of a predetermined size based on the center point. Method for calculating the fat content of meat from the ultrasound image, characterized in that. delete

Images