JPH0991418A - Image information processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To process an image at a high speed even when number of referred vicinity picture elements is increased by ranking plural picture element data in an observation window by quantity and using average data of the picture element data at a rank with many number of times of occurrence for picture element data. SOLUTION: A bit arithmetic means 4 receives plural picture element data for an observation window whose noise is eliminated and ranks the picture element data depending on the quantity into plural ranks, then each stage adder means 5 counts number of times of occurrence of each rank. Then a maximum value calculation means 6 selects and extracts a ranking occurrence most frequency among each ranking count of each stage adder means 5. A mean value calculation means 7 calculates a mean value of plural picture element value data corresponding to the rank extracted by the maximum value calculation means 6 and provides an output of data whose noise is eliminated. The effect of noise caused in a discrete way is easily eliminated by classifying the input picture element data into plural ranks of a few number and calculating the mean value of the picture element data of a ranking occurred most frequently in the ranks and the processing time is reduced.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image information processing apparatus for removing a noise component in image information at high speed when processing information of a digital image.

[0002]

2. Description of the Related Art Typical methods for reducing noise generated in a digital image include a method using an average filter and a method using a median filter.

The method using the average filter is the target pixel f
An average value of (x, y) (x, y are coordinates) and pixel values in the vicinity thereof is used as a pixel value after filtering. When expressed by a mathematical formula, g (m, n) ＝ (1 / N) ΣΣf (m + i, n + j). N is the number of neighboring pixels that have been added, and i and j are, for example, when a 3 × 3 window is selected as the observation window of the filter, -1,
It becomes 0,1 and becomes -2, -1,0,1,2 when a 5x5 window is selected. The above calculation is performed for all pixels to obtain an average image of all pixel values. With this processing method, the edges of the image are blurred. For this reason, this method is mainly applied to images with smooth density changes.

This averaging filter has a short calculation processing time, but has a drawback that it is affected by noise because it outputs an average value including pulse-like noise components of neighboring pixels. For example, assuming that 8 out of 9 neighboring pixels have a pixel value of 1 and only 1 pixel has a pixel value of 100 due to pulse noise, the output value of the pixel of interest is 1 in the median filter but 108/9 in the average filter. = 12 and a different value. In this way, the average filter method is greatly affected by the pixel value of pulse noise, and normal image information cannot be processed only by the average filter.

The method using the median filter is that the pixel value array in the observation window of the image is rearranged in ascending order, and the data at the intermediate position of the array, that is, the median value (median) is used as the pixel value at the pixel of interest. Is the way to get. This method is a very effective method because it can easily remove pulse-like noise that occurs discretely. For example, select at least a 3 × 3 window as the observation window of the median filter, and set the median of 9 pixel values as the pixel value after filtering, but observe it to sufficiently remove the influence of discrete noise and improve reliability. The window may be 5 × 5 and the median of 25 pixel values may be taken.

[0006]

However, the median filter method has a drawback that it takes a long processing time as described below. That is, in order to obtain the median, a plurality of data are sorted in, for example, ascending order, but the task of sorting takes a considerably long time even if a computer is used, and the processing time increases in proportion to the square of the number of data. In other words, if the processing of the averaging filter is one step, in the case of the median filter, a maximum of 9 squared (= 81) steps in a 3 × 3 observation window, and the average number of steps when all pixels are processed is about half. It takes 40 steps. If you select a 5x5 window, the average (25
Squared) /2.apprxeq.310 steps, which is a disadvantage that it takes 300 times as long as the average filter processing.

Therefore, the problem to be solved by the present invention is to reduce the noise included in the pixel value of digital image data by eliminating the effect of pulse noise of the pixel value and increasing the observation window in the vicinity. It is an object of the present invention to realize a noise removal method for digital image data that can be processed at a relatively high speed even if the number of reference pixels is increased.

[0008]

In order to solve the above-mentioned problems, in the structure of the present invention, a plurality of pixel value data in the observation window to be subjected to noise removal are received, and a plurality of ranks are formed according to the size of the pixel value data value. By dividing, the value obtained by calculating the average value of a plurality of pixel value data of ranks in which the number of occurrences of each rank is high is used as the pixel data from which noise is removed. Accordingly, in the image information processing apparatus that removes the noise component of the digital image data, the noise removal of the digital image data that can be processed at high speed is realized without the influence of the pulse noise of the pixel value.

More specifically, a rank dividing means (bit arithmetic means 4) for receiving a plurality of pixel value data in the observation window to be subjected to noise removal and dividing the plurality of ranks according to the size of the pixel value data is provided. Rank extraction means (each step addition means 5) for separately counting the number of occurrences is provided, and rank extraction for selecting and extracting a rank having a high occurrence number from each rank count value of the rank counting means (each step addition means 5). Means (maximum value calculation means 6) is provided, and calculation means (average value calculation means) that outputs noise removal data obtained by calculating an average value of a plurality of pixel value data corresponding to the extraction rank of the rank extraction means (maximum value calculation means 6) 7) is provided. Here, as the rank extracting means (maximum value calculating means 6), first, each rank counting value of each rank counting means (each stage adding means 5) has the maximum value, or secondly, the number of times of occurrence is large. There is a means for extracting and outputting a plurality of ranks.

As described above, the present invention distributes the pixel values of the pixels existing in the observation window into a plurality of ranks and registers the most pixels in order to obtain the noise removal performance according to the median. In this method, the average value is calculated only for the pixels belonging to the selected rank, and the average value is output instead of the median to increase the speed. That is, digital image data is input by the image input device and stored in the input image memory. The following processing is performed on the pixel of interest and its neighboring pixels of this image. The pixel value of each pixel is divided into a plurality of ranks by a threshold value. After ranking all the pixels, the rank in which the most pixels are registered is selected. The average value of only the pixels registered in the selected rank is calculated and used as the pixel value after filtering. When there are a plurality of ranks in which the largest number of pixels are registered, the average value of pixel values in all the plurality of ranks is calculated. By this processing, the pulsed noise is excluded from the processing for obtaining the average value, so that the calculation result can obtain a good result according to the processing by the median filter.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to examples.

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 is a block diagram showing the configuration of an embodiment of noise removal of digital image data according to the present invention, and FIG. 3 is a flow chart of the process described below. In FIG.
For example, digital image information obtained by A / D converting a screen imaged by a CCD camera is input from the input terminal 1 through the digital image input device 2 and stored in the input image memory 3 for each pixel f (x, y) at each point. The input image memory 3 stores the pixel value data of each pixel on one screen in the arrangement of the X axis and the Y axis.

As shown in this processing conceptual diagram of FIG. 2, the bit computing means 4 receives the pixel value data of the observation window unit of the input image memory 3, and 5 × 5 for each pixel f (m, n). The data of the neighboring pixels of is bit-shifted to the right to perform rank classification processing. In this example, each pixel shifts 8-bit (256 gradation) data to the right by 6 bits to create temporary data in which only the upper 2 bits are left. By this process, 0 to
Those with a pixel value of 63 have a value of 0, 64-127 have a value of 1,
Temporary data is created by converting 128 to 191 into a value 2 and 192 to 255 into a value 3.

In each stage addition means 5, the bit operation means 4 receives temporary data in which 5 × 5 = 25 pieces of data are divided into four levels, and counts the number of occurrences of each of the four ranks. That is, arrays m [0] to m [3] having subscripts 0 to 3 are prepared on the memory, and the contents of the array where the subscripts match the temporary data calculated by the bit operation means 4 are incremented by one. To count. The original 8-bit pixel value data is accumulated in the sum arrays p [0] to p [3] for the same rank to obtain the sum of the pixel values of each rank.

The maximum value calculation means 6 calculates the target rank based on the array data obtained by the stepwise addition means 5.
That is, the maximum array number of the count value, that is, the most frequent rank is taken out from the arrays m [0] to m [3]. However, when there are a plurality of plural count values, these plural array numbers are supplied to the average value calculating means 7 as the most frequent rank.

The average value calculating means 7 receives the maximum array element number of the most frequent rank obtained by the maximum value calculating means 6 and the sum array p [0] to p [3] corresponding thereto, and receives the target pixel value. The average value of is calculated and output to the output image memory 8. However, when the maximum value calculation means 6 has a plurality of sequence numbers,
Corresponding to this, the average value of a plurality of arrays is calculated and output.

The above-mentioned noise removal processing is sequentially carried out in the observation window unit over the entire pixel area, and after being stored in the output image memory 8, it is output through the digital image information output device 9 and displayed on the display device, for example. As described above, the input pixel value data is shifted and classified into a small number of multiple ranks, the pixel value data of the rank with the highest number of occurrences in that rank is noted, and the average value is calculated to obtain a discrete value. It is possible to easily remove the influence of noise generated in the above, and it is possible to significantly reduce the processing time because the processing time required by the conventional median filter method is not required.

In the above arrangement, the bit operation means 4,
Each stage adding means 5, maximum value calculating means 6, and average value calculating means 7 may be configured by a circuit, may be processed by a high-speed DSP, or may be realized by software processing by a CPU. .

[0019]

Since the present invention is configured as described above, it has the following effects. By classifying the input pixel values that should be subjected to noise removal processing into multiple ranks and calculating the average value only for the pixel value data of the ranks that have the highest number of occurrences, noise removal performance equivalent to that of median filtering processing can be obtained. An image information processing apparatus capable of easily removing the influence of discretely generated noise and capable of significantly reducing the processing time has been realized.

For example, when a median filter is applied to a 5 × 5 observation window, the processing time is about 310 times longer than that of the average filter method, but this method is slightly less than 30 times (5
Noise can be removed at a processing speed of (work for allocating × 5 pixels + work for finding maximum value + work for finding average value). In the median filter, the processing speed is proportional to the square of the number of pixels in the observation window, whereas in the present method, the processing speed is almost proportional to the number of pixels, and the time saving effect is great.

[Brief description of drawings]

FIG. 1 is a configuration block diagram of noise removal of digital image data according to the present invention.

FIG. 2 is a processing conceptual diagram for explaining a bit operation means 4 of the present invention.

FIG. 3 is a flowchart illustrating a procedure for removing noise from digital image data according to the present invention.

[Explanation of symbols]

 2 digital image input device 3 input image memory 4 bit computing means 5 each stage adding means 6 maximum value calculating means 7 average value calculating means 8 output image memory 9 digital image information output device

Claims (3)

[Claims] 1. An image information processing apparatus for removing a noise component of digital image data, receiving a plurality of pixel value data in an observation window to be subjected to noise removal, and dividing into a plurality of ranks according to the size of the pixel value data value,
An image information processing apparatus, wherein a value obtained by calculating an average value of a plurality of pixel value data of ranks having a high number of occurrences of each rank is used as pixel data from which noise is removed. 2. An image information processing apparatus for removing a noise component of digital image data, receiving a plurality of pixel value data in an observation window to be noise-removed, and classifying into a plurality of ranks according to the size of the pixel value data. Means is provided, rank-by-rank counting means for counting the number of occurrences for each rank is provided, and rank-extracting means for selecting and extracting a rank with a high occurrence frequency from each rank count value of the rank-by-rank counting means is provided. An image information processing apparatus comprising: a calculation unit that outputs noise removal data obtained by calculating an average value of a plurality of pixel value data corresponding to the extraction rank of the unit, and including the above. 3. The image information processing apparatus according to claim 2, wherein the rank extracting means extracts and outputs a rank having a maximum count value for each rank of the counting means or a plurality of ranks having a large number of occurrences.