JPS603671B2 - High-speed local parallel processing device for grayscale images - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a device for converting given image information into new image information.

Generally, image information obtained from an image input device is black and white.
It relates to grayscale images that include not only values but also intermediate tones.

This image information is used to numerically measure feature quantities, and before being calculated and judged, noise is removed, contours are extracted, and 2
It must be converted into the required new image information through processing such as digitization. Conventionally used means for this purpose are: (1) using a general-purpose computer, (2) using a microprocessor under microprogram control, switching the microprogram as necessary, and installing hardware corresponding to each process. They prepared them and changed them as needed.

However, these methods require a huge amount of memory in the computer (compared to
11, {Regarding 21) It had drawbacks such as the hardware being extremely large-scale (Regarding ■). An object of the present invention is to provide a processing device as a means for compressing image information obtained from an image input device and obtaining new image information remotely and economically.

The present invention will be explained below based on the drawings.

Figure 1 shows the central pixel 1 to be converted and its 8 neighboring pixels A, B, C, D, E, F, G when the screen is divided into pixels.
, a small region consisting of days.

If the screen is made up of square grid points, the minimum area considering isomerism is 3× consisting of the central pixel and 8 neighboring pixels.
Since it is known that there will be 3 areas, the present invention
Treat the region as a small region. Hereinafter, as shown in FIG. 1, the central pixel to be converted is indicated by 1, and its eight neighboring pixels are indicated by A, B, C, D, E, F, G, and day in sequence.
In addition, the density information of each pixel is indicated by a lowercase letter (a,
b, c, d, e, f, g, h, i converted pixels are distinguished by adding a dash. FIG. 2 is an embodiment showing a processing circuit configuration of the present invention.

Terminals I to Mako Kawa are input terminals of the circuit; density information i of the central pixel 1 is sequentially input to the terminals I, and density information a to h of 8 neighboring pixels A to day are sequentially input to the terminals I to H. .

Data selectors 1 to 8 have two input terminals, and select and output one of the two inputs according to an external command. This external instruction is inputted from the external terminal u of the selector 1 as shown in FIG. The two inputs of each data selector have the relationship shown in Table 1. Table 1 The first layer arithmetic modules 11 to 18 are composed of eight two-input, one-output arithmetic modules of the same type.

The inputs of each arithmetic module 11-18 have the relationships shown in Table 2, and each arithmetic module 11-18 executes an externally specified arithmetic operation between two inputs and outputs the result. One of the arithmetic modules 11 to 18 will be explained in detail with reference to FIG. 3, taking the arithmetic module 11 as an example. The calculation module 11 inputs the output of the selector 1 and the image information a from the pixel A, and performs basic calculations (shown in Table 3) between the two inputs. It is composed of a first modifier 71 made of, for example, TI's SN7 87N and SN7428, which respectively calculate and output the number of 2 buckets of the calculation result and the absolute value of the basic calculation result. Each calculation function of the basic calculation circuit 61 and the first modifier 71 is designated by a control signal input from the terminal u. Table 2 1st
Each calculation module of the layer calculation module and its input table 3 Calculation functions when the two inputs of the calculation module are x and y ff = x, f = y, f = x + y, f = x - y, f=×
-y-1 The second layer arithmetic modules 21 to 24 are composed of four 2-input, 1-output arithmetic modules of the same type.

The inputs of each arithmetic module 21-24 have the relationships shown in Table 4, and each arithmetic module 21-24 executes an externally instructed arithmetic operation between two inputs and outputs the result. One of the calculation modules 21 to 24 will be explained in detail with reference to FIG. 4, taking the calculation module 21 as an example. The calculation module 21 receives the outputs of the calculation modules 11 and 18 of the first layer calculation module as inputs, and includes a basic calculation circuit 61 that performs basic calculations (shown in Table 3) between the two inputs;
For example, TI's SN74S performs calculations such as selecting the larger or smaller of two inputs and outputting the contents of the latch 25, which can arbitrarily set a constant from the outside.
It consists of a second modifier 81 made in I5. Each calculation function of the basic calculation circuit 61 and the second modifier 81 is specified by a control signal input from the terminal u. Table 4 As shown in Figure 2, the calculation modules 21, 23
The latch 25 provided in the calculation module 22 and the latch 26 provided in the calculation modules 22 and 24 can each have independent constants set.

The third layer arithmetic modules 31 and 32 are composed of two 2-input, 1-output arithmetic modules of the same type.

The inputs of each arithmetic module 31, 32 have the relationship shown in Table 5, and each arithmetic module 31, 32 executes an externally specified arithmetic operation between two inputs and outputs the result. One of the calculation modules 31 and 32 will be explained in detail with reference to FIG. 5, taking the calculation module 31 as an example. The calculation module 31 is the calculation module 22 of the second layer calculation module.
, 24 as inputs, and a basic calculation circuit 61 that performs basic calculations (shown in Table 3) between the two inputs, the absolute value of the basic calculation result, and selection of the larger or smaller of the two inputs. For example, SN7 Nagi87N, SN742
It consists of the third modifier 91 made in Europe, SN7 semi-157N.

Each calculation function of the basic calculation circuit 61 and the third modifier 91 is designated by a control signal input from the terminal u. Table 5 The fourth layer arithmetic module 41 is a two-input, one-output arithmetic module, which receives the outputs of the third layer arithmetic modules 31 and 32 as input, and executes arithmetic operations between both inputs in response to instructions from the outside.

The fourth layer calculation module 41 will be explained in detail with reference to FIG. The fourth layer calculation module 41 includes a basic calculation circuit 61 that has two inputs and performs basic calculations (shown in Table 3);
For example, TI's SN7 Yama8 which calculates and outputs the absolute value of the basic calculation result, the selection of the larger or smaller of two inputs, etc.
A third modifier 91 made of 7N and SN7428, and a third modifier 91 made of SN7428, for example TI's SN7 standard, which takes the output of the third modifier 91 as input and has arithmetic functions such as multiplying the input by two rivers (n is an integer). The division circuit 101 made of 157N and the calculation module 31 of the third layer calculation module
, each sign bit output S of the 32 basic arithmetic circuits,
A binarization circuit 12 that has an arithmetic function that uses the contents of the sign bit of S2 to output 0 or 1 according to set conditions.
1, and an arithmetic function for selectively outputting either the output of the division circuit 101 or the output of the binarization circuit 121, for example, the SN7 So 157.
It is composed of a selector 111 made from N.
and a basic arithmetic circuit 61, a third modifier 91,
Each calculation function of the division circuit 101, the binarization circuit 121, and the selector 111 is designated by a control signal input from the terminal u. The fifth layer arithmetic module 51 is a two-input, one-output arithmetic module, and receives the outputs of the center pixel 1 and the fourth layer arithmetic module 41 as inputs, and outputs the output of the fourth layer arithmetic module 41 as is, or outputs the output of the fourth layer arithmetic module 41 as it is, or It has a function to compare and select and output the larger or smaller one.

More specifically, as shown in Figure 7, for example, SN 700 million
The selector 141 made of 57N has an arithmetic function that selects and outputs the output of the center pixel 1 and the fourth layer arithmetic module 41, while the comparison circuit 131 made of, for example, TI's SN74S8 arc selects and outputs the output of the center pixel 1 and the fourth layer arithmetic module 41. Arithmetic module 41
It has an arithmetic function that compares the outputs of and sends a signal to the selector 141 to select and output the larger or smaller one. The selector 141 and the comparison circuit 131 constitute the fifth layer arithmetic module 51, and each arithmetic function is designated by a control signal input from the terminal u. With this circuit configuration, input selectors 1 to 8, each calculation module 11 to 18, 21 to 24, 31, 32, 41, 5
By specifying the function of 1 and the contents of the latches 25 and 26 from the outside through the terminal u of each circuit, it is possible to realize a wide variety of filters with regard to the operation order and contents that are compatible, as shown in Table 6, for example. .

Table 6 Noise removal i' = (a + b ten c ten d ten e ten f ten g ten h)
/ji contour extraction i'=max(la-i l, lb-i
l, lc-il, ld-il, le-・l, lf-i
l, lg-il, lh-il) Binarization i'=l if-》i>L, ratio lsl average gradient i'=(la-il 10lb-il 10lc-il+l
d-il ten le-il+! f-il ten lg-il ten lh
-1 l)/is the difference in the x direction i'=l(a+g+h)-(c+d+e)
lY direction difference i'=l(a+b+c)-(e+f×g
) l difference r = {l (a + b + c) - (e + f + g) l
-l(a0g+h)-(e+d+e)l}/2 Laplacian i'= {18i-(a+b0c+d+
e+f+g+h)} Noji 4-way Laplacian i′{4
i - (b + d + f + h)} / range filter i' max (a, b, C, d, e, f, g, h, i) - mi
n (a, b, c, d, e, f, g, h, i) The operation of the embodiment shown in Fig. 2 will be explained below, taking noise removal, contour extraction, binarization, and X-direction difference as examples. Explain in detail.

'1' Noise Removal Filter If we assume that the noise is an 8-function noise, we can calculate the average density of pixels in a small area containing that noise and replace it with the center pixel, which will give us the image information from which the noise has been removed. is obtained.

In this case, the functions of all the first layer arithmetic modules 11 to 18 are externally designated so as to output inputs other than those from the selector as they are.

As a result, their outputs are a, b, c, d, e, f in order
, g, h. The second layer calculation modules 21 to 24 are 2
The function is specified externally to output the sum of inputs. As a result, the outputs are h0a, b+c,
d+e, f+g. Third layer calculation module 31, 3
The function of 2 is externally designated to output the sum of two inputs. As a result, the output is b + c + f
10g, a+h10d+e. 4th layer calculation module 4
1, the function is externally designated to output the sum of two inputs and a value shifted by 3 bits. As a result, the output is (a+b+c+d+e+f+g+h)/8. The function of the fifth layer arithmetic module 51 is specified from the outside so that it outputs the output of the fourth layer arithmetic module as it is. As a result, the density information (a+b+c+d+h)/8 is outputted to the output terminal of the fifth layer calculation module 51 as the new value i' of the central pixel to be converted. ■ Contour line extraction filter Considering that the contour line is the boundary between the object image and the background, it is a part where the density changes suddenly, and by calculating the density difference between pixels in a small area and extracting the part where this difference is large, the contour line can be extracted. Image information showing the image can be obtained.

The functions of all the selectors 1 to 8 are externally specified so as to select the density information i of the center pixel 1. As a result, the inputs of the first layer calculation modules 11 to 18 are (i, a) in order.
, (i, b), (i, e), (i, d), (i, e),
(i, f), (i, g), (i, h). The first layer arithmetic modules 11 to 18 specify their arithmetic functions from the outside so as to output the absolute value of the difference between two inputs. As a result,
The outputs are li-al, li-bl, li-cl,
li-dl, li-el, li-fl, li-gl, l
It becomes i-hl. The second layer calculation modules 21 to 24 are 2
The arithmetic function is specified externally to compare the inputs and select and output the larger one. As a result, its output is m
aX(l i-a l, l i-hl), maX(l
i-cl, l i-bl), max(l i-e l,
li-dl), max(li-gl, li-fl). The function of the third layer calculation modules 31 and 32 is specified from the outside to compare two inputs and select and output the larger one. As a result, the output is max(li-Cl
, lj-bl, li-gl, li-fl) maX(li
-al, li-hl, li-el, li-dl).

The function of the fourth layer arithmetic module 41 is specified from the outside to compare two inputs and select and output the larger one. As a result, the output is max(li-al, li-bl, l
i-Cl, li-dl, li-el, li-fl, li
-gl, li-hl).

In the fifth layer calculation module 51, the fourth layer calculation module 4
The function is specified from outside so that the output of 1 is output as is. As a result, the output terminals of the fifth layer calculation module 51 have max(li-al, li-bl, li-Cl
, li-dl, li-el, li-fl, li-gl,
li-hl) is the new value i to be converted for the central pixel.
′ is output.

Furthermore, if necessary, only the outline will appear on the screen by performing binarization, which will be described later.

8} Binarized image information can be obtained by comparing the density of each pixel with the reference density given for each binarization filter and replacing the value of each pixel with 1 or 0 based on the result.

The functions of all the selectors 1 to 8 are externally specified so as to select the density information i of the center pixel 1.

The first layer calculation modules 11 to 18 output the input from the selector side as is. The latches 25 and 26 are given an appropriate constant 12, 1, 02>1,) in advance, and the second
Among the layer calculation modules, calculation modules 21 and 22 output constants of latches 25 and 26, and calculation module 23 outputs constants of latches 25 and 26.
, 24, the function is externally designated to output one of the two inputs as is. As a result, the outputs of the second layer calculation modules 21 to 24 become 12, 1, 1, and i in this order. The functions of the third layer calculation modules 31 and 32 are specified from the outside so as to take the difference between two inputs. Here, the arithmetic module 31 executes 12-i, and the arithmetic module 32 executes 1, -i, and the sign bit becomes 1 only when the result is negative. The fourth layer arithmetic module 41 uses the two sign bits obtained from the third layer arithmetic module and reads 1 only when the sign bit from the arithmetic module 31 is 0 and the sign bit from the arithmetic module 32 is 1. The arithmetic function is specified externally to output.
The fifth layer arithmetic module 51 has its arithmetic function specified from the outside so that it outputs the output of the fourth layer arithmetic module as it is. As a result, the output terminal of the fifth layer arithmetic module, which is the output terminal of the processing circuit, outputs 1 if the density i of the central pixel is 12≧i>1 for constants 1, 12, and 0 otherwise. Then, a binarized filter is obtained. '4}
Difference in X Direction In FIG. 1, if the direction of pixel ○ from the pixel date is defined as the x direction, the difference in the x direction is given as the absolute value of the difference in density information (d+h+g) and (e+d+c). The functions of all selectors 1 to 8 are externally designated to select density information a, b, c, d, e, f, g, and h from pixels other than the center pixel 1.

As a result, the inputs of the first layer calculation modules 11 to 18 are (h, a), (a, b), (b, c), (C, d),
(d, e), (e, f), (g, h). In the first layer arithmetic module, the arithmetic modules 11 and 15 each output the input other than the selector as it is, and the arithmetic modules 14 and 18 specify their arithmetic function from the outside so as to output the sum of two inputs. As a result, the outputs of the calculation modules 11, 14, 15, 18 are (a),
(c+d), (e), (g+h). In the second layer arithmetic module, each of the arithmetic modules 21 and 23 specifies its arithmetic function from the outside so as to output the sum of two inputs. As a result, the outputs of the calculation modules 21 and 23 become (a+g+h) and (c+d+e) in this order. In the third layer arithmetic module, the arithmetic module 32 externally specifies its arithmetic function so as to output the absolute value of the difference between two inputs. As a result, the output of the calculation module 32 is l
(a0g1h)-(c+d+e)l. The calculation functions are specified from the outside so that the fourth layer calculation module 41 outputs the output of the calculation module 32 as it is, and the fifth layer calculation module 51 outputs the output of the fourth layer calculation module 41 as it is. As a result, the output terminal of the fifth layer calculation module 51 is l(a+g+h)−(c+d+
e) Intensity information l is output as the new value i' of the central pixel to be transformed. In the above four types of examples, the fifth layer calculation module 51 outputs the result of the fourth layer calculation module 41 as it is, but when executing the range filter (shown in Table 6) as the filter, The fifth layer calculation module 51 selects the maximum value and minimum value from the density information from nine pixels. In this case, in addition to the memory that stores the given image data, a memory that can store new image data that has been processed is required.

Furthermore, a device is required that can measure the difference between pixels at the same position within the screens of two images. In addition, in practice, when executing other filters shown in Table 6, if you prepare a memory to store the given image data and a memory to store the new processed image data, this device can be used. It can be used more effectively. Note that when the selector selects and outputs image information from 8 neighboring pixels, it is only when executing the X-direction difference, the Y-direction difference, and the difference, so if these three filters are not required, the selector is not necessary.

Furthermore, in the present invention, the input and output relationships between the selector and the calculation modules of each layer are not limited to the present embodiment shown in FIG. Even if you connect the modules as shown below, many filtering results will not change. In other words, (11, 1
2), (13, 14) “(15, 16), (17, 18
) may be formed as inputs to the third layer calculation modules 21, 22, 23, and 24, respectively.

Further, although the minimum necessary basic calculation functions for the present invention are shown in Table 3, these basic calculation functions are not limited to those in Table 3, and the components within each calculation module may be appropriately selected. Therefore, the filters shown in Table 6 can also include filters other than those shown.

According to the processing device of the present invention described above, there is no need to prepare hardware corresponding to each process, and by simply changing the function of the arithmetic module and the data passage route, multiple pieces of hardware corresponding to each process can be prepared. Different types of filters are configured.

Additionally, if the input of this device is connected to an image memory that continuously outputs image information from small areas within the screen,
By once specifying the function of each calculation module, new compressed image information for each pixel can be obtained one after another using only the propagation delay time in the processing element. There is no need for the addressing and readout time of the program, which is required when performing calculations, and the addressing time required for reading each pixel, and the processing time is extremely shortened.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing the names of each pixel in a 3×3 area, FIG. 2 is a configuration diagram of an embodiment of the present invention, and FIG. 3 is a selector and one calculation module 11 of the first layer calculation module. 4 is a detailed configuration diagram of one calculation module 21 of the second layer calculation module, FIG. 5 is a detailed configuration diagram of one calculation module 31 of the third layer calculation module, and FIG. The figure shows a detailed configuration diagram of the fourth layer calculation module 41,
FIG. 7 is a detailed configuration diagram of the fifth layer calculation module 51. [Explanation of symbols of main parts], A to I; Input of processing circuit,
1-8: Selector, 11-18; 1st layer calculation module, 21-24; 2nd layer calculation module, 31-32; 3rd layer calculation module, 41; 4th layer calculation module, 5
1; 5th layer calculation module, 1: central pixel, A, B, C
, D, E, F, G, H: 8 neighboring pixels, i: Image information. spear 1 illustration spear 2 illustration uta 3 illustration spear 4 illustration spear 5 illustration spear 6 illustration spear 7 illustration

Claims (1)

[Claims] 1. A device that converts a central pixel into desired new image information using image information from the central pixel and eight neighboring pixels, capable of transmitting at least image information from the central pixel. A transmission means, and a two-input device that inputs image information of the transmission means from one input terminal and inputs image information of a specific neighboring pixel from the other input terminal, and performs an operation between the two.
A first layer arithmetic module including a total of eight output arithmetic modules corresponding to each neighboring pixel, and one of four pairs consisting of eight outputs of the first layer arithmetic module is used as an input to perform arithmetic operations. A computation module with one input and one output is associated with each pair, and a computation is performed using one of the two pairs as input, which consists of a second-layer computation module containing four in total, and four outputs of the second-layer computation module. a third layer calculation module that includes a total of two 2-input 1-output calculation modules that correspond to each pair; and a fourth layer that includes a 2-input 1-output calculation module that performs calculations between the outputs of the third-layer calculation modules. a layer calculation module; and a control means for specifying the calculation function of the calculation module included in at least the first layer calculation module, the second layer calculation module, the third layer calculation module, and the fourth layer calculation module, respectively. A high-speed local parallel processing device for gray scale images. 2. In the processing device according to claim 1, the transmission means includes eight selectors with two inputs and one output corresponding to each neighboring pixel, and one input terminal of each selector receives an image from the central pixel. The information is configured such that image information from a specific neighboring pixel is input to the other input terminal, and
The output terminal is connected to one input terminal of the arithmetic module that uses a pixel adjacent to a specific neighboring pixel as specific pixel information, and causes the control means to output one of the two inputs of the selector. 1. A high-speed local parallel processing device for grayscale images, characterized in that a function for specifying an arithmetic function of the selector is added.