JPH05165963A - Extraction of image contour - Google Patents

Abstract

PURPOSE:To extract a contour in real time by utilizing a signal generated by subtracting a detail component of an image from an image edge component and also using the absolute value of the secondary differential component of input image data as a detail component of the image. CONSTITUTION:The signal generated by subtracting the detail component of the image from the image edge component is used for target contour information so as to separate a noise at an edge periphery without using fine adjustments for threshold value setting in a threshold value process which extracts a contour line indicating the border of a body in the image. Further, the absolute value of the secondary component of the input image data which is obtained by an absolute value means 11 is used as the detail component to perform the threshold process which has a sufficient margin for the noise at the periphery of the edge by contour extraction. Consequently, a binary signal which is necessary as border information can be extracted by correcting a threshold value with the intensity distribution of the obtained signal value. Consequently, the contour extracting process can be performed in real time.

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 means and a block coding means required for a compression coding method of multi-valued digital moving image data, and a contour line of an image in the fields such as computer vision. The present invention relates to extracting the contour of an image that can be used when a non-rectangular region divided by is used as a processing unit.

[0002]

2. Description of the Related Art A number of precedents have been shown in the field of image recognition and the like in an attempt to extract contour line information indicating an object boundary from an image. In these known methods, the output of a spatial filter that emphasizes edges is binarized by a certain threshold. However, it is necessary to manipulate the threshold according to the image condition to obtain the desired result. Therefore, select parameters such as thresholds that are optimal for specific scenes from empirical values, or use
Alternatively, when a natural image in which various scenes are expected to occur is targeted, image analysis such as auxiliary histogram analysis for determining processing parameters is required, so it is necessary to add an interactive operation of the operator. there were.

Generation of a binary contour signal for the purpose of performing block processing based on boundary information such as a contour line in an image in image compression coding or the like is more difficult than in the past because of a limitation of processing time. It is not possible to determine the parameters in the way they are done. In order to configure a means for processing in real time or quasi-real time, it is preferable that the optimum parameters can be automatically determined by a simple method, and its binary output indicates a boundary, and its output is approximately It must be binary information indicating an image with a line of constant width.

[0004]

If a simple threshold value process is used when extracting a binary contour image, it is necessary to adjust parameters such as a filter and threshold value to extract a target contour line. That is, it is difficult to realize high-speed processing in real time.

[0005]

According to the present invention, there is provided thresholding processing for extracting a contour line representing a boundary of an object in an image,
A signal obtained by subtracting the fine component of the image from the image edge component is used in order to make it possible to separate the noise around the edge from the target contour information without using fine adjustment for the threshold setting. Further, by using the absolute value of the second-order differential component of the input image data as the fine component of the image, extraction as a contour enables threshold processing with a sufficient margin against noise around the edge. By correcting the threshold value based on the intensity distribution of the signal values obtained as a result, it becomes possible to extract the binary signal required as the boundary information without being affected by the intensity of the image edge. By performing these processes, the contour extraction process can be realized in real time.

[0006]

FIG. 1 is a block diagram showing an embodiment of the device of the present invention. Image edges are transferred via block delay elements 2 to 4 which delay data read out by raster scan (scan line sequential) read from the frame memory 1 storing multi-valued input image data 19 for one screen by one line. The first spatial filter 8 for detecting, and the second spatial filter 8 for detecting fine components of the image through the block delay elements 4 to 7 as well.
The spatial filter 9 is provided, and the difference between the outputs of these two filters is obtained by the subtracting means 10 and the frame memory 12 for delaying by one frame via the absolute value means 11 for outputting the absolute value and one frame for output. The average value is calculated from the data and is input to the average value calculation holding means 13 which holds the value by the frame reset signal 21.

A multi-tap block delay element 15 in which delay elements for one line of a plurality of lines which require data read from the frame memory 12 are cascaded
It is input to the median filter 16 for obtaining the median value of the neighboring pixels via the, and from the output of the median filter and the output of the average value calculation holding means 13, the threshold value determining means 17 outputs 2
The threshold of binarization is determined and the output of the frame memory 12 is input to the threshold discriminator 18 via the block delay element 14 to generate binary contour information and output as an output signal 20. As a specific example of the spatial filter, a Robert gradient calculator or a Sobel calculator is used for the first spatial filter 8, and a high-pass filter having a filter coefficient for an adjacent pixel is used for the second spatial filter 9. It can be realized by using a filter. In the third aspect, the second spatial filter 9 is a cascade connection of the high-pass filter and a filter that outputs an absolute value.

FIG. 2 is a diagram for explaining the operation of the means realized in FIG. 1, and it is assumed that image data of the same value is processed in the vertical direction. Assuming that the image data read from the frame memory 1 of FIG. 1 has a value represented by the waveform of FIG. 2-a, the edge detection output of the first spatial filter 8 and the high frequency band of the second spatial filter 9 are high. The output of the pass filter has waveforms as shown in FIGS. 2B and 2C, respectively. Here, the edge detection of the first spatial filter 8 outputs a value represented by the following equation.

S = {(Sx) ² + (Sy) ² } ^1/2 / 8 where Sx = (I3 + 2I6 + I9)-(I1 + 2I4 + I7) Sy = (I7 + 2I8 + I9)-(I1 + 2I2 + I3)

In the above equation, Sx and Sy represent local changes (gradients) in the respective directions of the square lattice (FIG. 4) which constitutes the image sample. S is a value representing the strength of the gradient. I1 to I4 and I6 to I9 represent the values of the pixels adjacent to the edge detection position.

FIG. 2-d is the subtraction of the absolute value of FIG. 2-c from FIG. 2-b, which corresponds to the output of 10 of FIG. For this reason,
When this waveform is used for the contour extraction processing, it is possible to obtain a result in which the threshold value may be roughly determined as compared with the case where FIG. 2B is used for the processing.

By dynamically determining the threshold value from the screen average value and maximum value of the waveform of FIG. 2D, edge information that is difficult to detect when binarized with a single threshold value is detected. It becomes possible to do. 1 obtained by the above process
An average value is calculated for the screen data and used as a threshold value. The multilevel contour signal generated by this is
When a natural image is used as the original data, a signal distribution represented by FIG. 5 is shown. In FIG. 5, Md represents the median value, MA represents the average value, and L represents the maximum value. By using the MA in FIG. 5 as a threshold value, it is possible to easily configure a means for extracting contour information similar to human visual characteristics.

Further, when the ratio of pixels that locally represent an edge exceeds a certain level, it is often not binary data suitable as contour information, and threshold correction is required. If the median value in the vicinity of the pixel to be processed is used as the threshold value, the generation ratio of binary values in the area can be set to about 50%, and the means for solving the above problems can be achieved.

[0014]

According to the present invention described above, a binary image suitable for use as boundary information in an image can be generated with a relatively simple structure.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 is an explanatory diagram of a waveform of each part.

FIG. 3 is a diagram in which filter coefficients are displayed according to sample positions on a screen.

FIG. 4 shows the positions of pixels used for edge detection and their values.

FIG. 5 is a diagram showing an example of a frequency distribution diagram of multilevel contour signals obtained as a result of signal processing according to the embodiment.

[Explanation of symbols]

 1, 12 Frame memory 2 to 7, 14 Block delay element 8 First spatial filter 9 Second spatial filter 10 Subtracting means 11 Absolute value means 13 Average value calculation holding means 15 Multi-tap block delay element 16 Median filter 17 Threshold Value determining means 18 Threshold discriminator 19 Input image data 20 Output signal 21 Frame reset signal

Claims (3)

[Claims] 1. A means for generating a binary contour image from a multivalued digital image, wherein a first spatial filter for extracting an image edge from multivalued image data which is an original signal and a fine component of an image from the original signal are extracted. A second spatial filter that
A multivalued contour signal is generated through a difference means for generating a difference between the output signals of the first and second spatial filters and a means for obtaining an absolute value of the difference output, and threshold processing is performed to generate a contour Image contour extraction, comprising means for determining a binary signal indicating presence / absence and dynamically determining a threshold value based on an average value of a screen of generated signals. 2. A median value in a rectangular area block including the pixel of the same signal, where MA is a screen average value of multivalued contour information.
The image contour extraction according to claim 1, further comprising a threshold discriminating means for setting a threshold value to MI when MI is larger than MA, and a threshold value to MA in other cases. .. 3. A high-pass filter for calculating a second-order differential coefficient is used as a second spatial filter for extracting a fine component,
The image contour extraction according to claim 1, further comprising outputting the absolute value.