JPH01175083A - Image processor and method of image processing - Google Patents

Abstract

PURPOSE:To effectively extract an edge without extracting noise other than an edge part by executing edge extracting gradation conversion prior to edge extraction. CONSTITUTION:The edge extraction of an image memory 10 is executed by using a differential operator shown in Fig. (a) and the extracted result is stored in a work image memory 14. Then, OR operation between the contents of the memory 14 and a work image memory 13 is executed and the result is stored in an image memory 13. Similar operation is executed by using an operator obtained by counterclockwisely rotating the Fig. (a) like Fig. (b). Thereafter, said processing is repeated in all eight directions. Finally, the gradation conver sion of the data in the image memories 10-12 is executed by using lookup tables 17-19 in accordance with the succeeding using purpose.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to an image processing device.

[Conventional technology]

Conventionally, in image processing apparatuses and methods using edge extraction, it has been common to read a color image into a memory and directly perform edge extraction using a differential operator.

[Problem that the invention is trying to solve] However,
In the conventional example above, edge extraction is performed on the original image imported into the frame memory, the data as it is, so if you only operate the threshold value to obtain the edges, noise may be picked up in areas other than the edges. Although it is necessary as an edge, it has the disadvantage of not being able to detect edges.

In view of this drawback, it is an object of the present invention to provide an image processing method and apparatus that can perform edge extraction without picking up noise other than edge portions.

[Means for Solving the Problems (and Effects)] To achieve the above object, according to the present invention, gradation conversion for edge extraction is performed before edge extraction.

〔Example〕

According to the embodiment of the present invention described below, by providing a gradation conversion means for performing gradation conversion suitable for edge extraction before performing edge extraction, edge extraction is performed after performing gradation conversion. By performing edge extraction with low noise and good continuity, and by providing a gradation conversion means to return to the natural gradation after edge extraction, the gradation-converted image described above can be An image processing device that converts the gradation back to a gradation according to the purpose of use will be described.

Hereinafter, one embodiment of the present invention will be described in detail using the drawings. FIG. 1 shows the entire illustration production apparatus in one embodiment of the present invention.

In Figure 1, 1 is the image or object that is the original image of the illustration,
2 is an object imaging device, for example, a TV camera is used. Reference numeral 3 denotes an image processing section that processes the captured image, and the details of the processing will be described later.

Reference numeral 4 denotes an output device that outputs the results processed by the image processing section, for example, as a CRT monitor or a hard copy.

The illustration production apparatus configured as described above will be explained below according to the flow of data. In FIG. 1, an image of an object l, for example a person, on which an illustration is to be made is captured by an imaging device 2, for example a TV camera. The image signal is A/D converted and taken into the image processing section 3.

The results processed by the image processing unit 3 are output as a hard copy by an output device 4 such as a color video printer.

FIG. 2 is a block diagram showing the image processing section 3 in more detail. The CPU 5 is a control processor that controls the entire first drawing image processing section 3 together with the CPU memory 6. A tablet is connected to the parameter setting unit 1108, and coordinate information can be sent to the CPU 5. The image processor 9 is the core part of the image processing unit, and according to the instructions of the CPU 5, the results are stored in any selected image memory 10 to 15 or image data l1016.
send to Image memory lO~14 is 512X512X
It has a size of 8 bits. The 16-bit work memory 15 has a 16-bit structure in the depth direction, and is used for image operations using more than 8 bits, such as cumulative addition, multiplication, and division. Image memory 10-15 is CPUBU
Since they are connected by S, it is possible to read and write to and from any image memories 10 to 15 by commands from the CPU 5, and it is also possible to perform real-time calculations between arbitrary memories.

Exit of image memory 10 to 12 (readout IIJ)
has high-speed RAM with lookup tables 17 to 19.
is connected. This RAM has a 256 x 8 bit structure, with 8 RAM address lines (0 to 25
Address 5 (0 to 255 gradations can be specified) is directly connected to the output of each image memory, and 8 data lines are VIDE
Connected to OBUS. Also, the contents of RAM are
5 allows you to read and write freely.

9 and 10 show lookup tables 17 to 19.
This is a graph showing the state of , where the horizontal axis shows the input gradation and the vertical axis shows the output gradation. Figure 9 shows the standard state, with 0 at address 0.
．． It shows that the value 255 is written in the 1st address and 1,255th address, and the input and output contents are the same.

In addition, Fig. 10 shows 255° at address θ, 254 at address 1,
This shows a state in which a value O is written at address 255. When using the lookup table shown in FIG. 10, the relationship between input and output is an image in which the luminance data is inverted.

The TV camera 2 and output device 4 shown in FIG. 1 are connected to the image data I1016, and input and output image data according to instructions from the CPU 5.

FIG. 3 shows a flowchart of illustration processing (processing to obtain a portrait from a human face) using the edge extraction method of the image processing method of the present invention, which is performed in the first drawing image processing unit 3. The processing method will be explained in detail below according to the data flow. Here, for convenience, data of red, green, and blue components forming a digital image are expressed as R9G and B. Furthermore, one pixel is composed of 8 bits for each component, that is, data capable of expressing 256 gradations, with the highest brightness being 255 and the lowest brightness being 0.

Appropriate lighting is applied to the subject of the step S1 for one day, and once the composition is determined, information to start processing is sent from the tablet connected to the parameter setting l108. When the CPU 5 receives this information, it sends a TV right camera power command to the image data I1016, and as a result, the RGB luminance data quantized to 8 bits is stored separately in each image memory 10, 11, 12. The image memories 10, 11.12 correspond to RGB video signals in order, and the operator can check the status of the image memories 10, 11.12 on the CRT monitor.

fB of suit S2 Next, the image processor 9 performs edge extraction processing according to a command from the CPU 5. This part is a feature of the embodiment of the present invention. In the existing method, edges are extracted using, for example, a 3×3 differential operator, and then the edge portions are connected to create a closed region. However, in the above-mentioned method, after filtering processing, only binarization is performed using a certain threshold value, so edges are often cut off in the middle or edges are detected in unnecessary parts. Therefore, in the embodiment of the present invention, edge detection is performed after converting the gradation of the original image data using lookup tables 17 to 19 so that appropriate edges can be obtained in advance before edge detection. The above-mentioned problem was solved. The contents of the lookup table are experimentally determined according to the characteristics of the operator used for edge extraction, and are written into the CPU memory 6. FIG. 7 shows the contents of lookup tables 17 to 19 used in this embodiment. This table is used when edges occur in a relatively high brightness area, such as a human face, and a characteristic that stretches the high brightness area is selected. In this embodiment, the same table is used for lookup tables 17-19. Also, as this lookup table, for example, the luminance data is O~255.
A table may also be used in which a histogram of the frequency of occurrence is created, a cumulative histogram of this histogram is created, and tone conversion is performed using this cumulative histogram. See you at 17! Instead of using the same table for each color, a table may be created for each color. Image memory 10-12
The data is gradation-converted in real time and written to the image memories 10 to 12 again.

Next, edge extraction is performed using a 3×3 differential operator.

FIG. 5 shows three typical examples of differential operators.

First, the R component is processed. In other words, image memory 1
0 using the differential operator shown in FIG. 5(a), and the results are stored in the work image memory 14.
Store in. Then, an OR is performed with the work image memory 13 and the result is stored in the work image memory 13. Similarly, the same process is performed using the operator who rotates counterclockwise from FIG. 5(a) to FIG. 5(b). Similarly, processing is performed using FIG. 5(C), and processing is performed for all eight directions below.

When the processing is completed for the image memory IO, the image memory 11. Similar processing is performed on the image memory 12 and all ORs are taken. Then, binarization processing is performed at a predetermined level. As a result, the edges are connected, and the image is divided into several closed regions.Hereafter, each of these regions will be referred to as a uniform color region.

Finally, the data in the image memories 10-12 are subjected to gradation conversion using look-up tables 17-19 depending on the purpose of use thereafter. In other words, if the same data as that at the time of input is required, an inverse conversion table of the table shown in FIG. 7 may be used. In this embodiment, a table shown in FIG. 8 suitable for filling in closed areas is experimentally determined and used later.

The gradation-converted data is stored in the image memories 10 to 12 again.
is stored in

Figure 6 shows an example of the results obtained in step S2 when a human face is processed. In this case, images in the work memory 13 and image memories 10, 11, and 12 are combined. In this figure, the black parts are the edge parts, which generally have a high frequency range (and the white parts have a high frequency range). Also, Figure 4 shows the priority order of the screen displayed on the CRT monitor. In other words, since the work image memory 13 has a higher priority than the image memories 10, 11 and 12, it appears as if an outline has been drawn on the color image.

The image processor 9 receives instructions from the CPU 5 and labels each area surrounded by the outline. Labels are applied only to hair regions and facial skin regions. This labeling procedure will be explained. First, select two areas on the screen from the largest area, and select the one with the largest maximum value of Y for each area as the hair portion in FIG. 6, and the other as the skin portion in FIG. 6 (■). In the CPU memory 6, RGB values suitable for filling in hair and skin colors are recorded in advance.

When filling in, the impression that a person's skin color gives to the viewer is thick (and even if you fill in using the average value of the skin color part of the original image, it is usually not a color that is subjectively desirable as a skin color.) Therefore, preferred RGB values have been determined experimentally and are stored in the CPU memory 6.The image processor 9 receives instructions from the CPU 5, and processes the areas (■) and (■) in FIG. 6, respectively. Coloring is performed automatically using the values stored in the CPU memory.The images that are filled in here are image memories 10, 11, and 12.
These are the only hair and skin parts of the image that do not have bits set in the image memory 13.

In other words, only the white parts (low frequency parts) in FIG. 7 are filled in, and the original image information remains as is in the black outline parts (high frequency parts). As a result, high-frequency components such as the areas around the eyes and hair remain as the original image, while low-frequency components are filled in uniformly, creating a detailed illustration that looks as if it was drawn by a skilled illustrator using poster colors. It's done.

Erase the snow image memory 13 of suit S4. As a result, the outline seen on the CRT monitor disappears, and the illustration of the portrait is completed.

The contents of the 9th day image memory 10, 11.12 of the suite S5 are outputted by the video printer, and the process is completed.

In this embodiment, a three-dimensional object was directly photographed using a TV camera and the image was input, but the same effect can be obtained by inputting from a printed matter, film, or photographic print. Although a TV camera is used as the input device, the same effect can be obtained by inputting image data via a recording medium such as a still video floppy disk, video tape, magnetic disk, magnetic tape, or optical disk.

Further, in this embodiment, the same table is used for R, G, and IB data, but it is also possible to use separate tables for R, G, and H. In this case, for example, the aforementioned cumulative histograms are created for each of the R, G, and H data, and edges are detected after performing gradation conversion of each data. For color images,
This method is very effective because the histograms for each color vary greatly. Further, as the gradation conversion table for edge extraction, various tables other than the table shown in the embodiment can be used.

As explained above, according to this embodiment, by performing gradation conversion before edge extraction and then performing edge extraction,
Edges with good continuity and less noise can be obtained, and
By performing gradation conversion after edge extraction, it is possible to obtain gradation reproduction data suitable for later use, and it also saves memory because there is no need to save the original image at the time of input. There is an effect.

〔Effect of the invention〕

As described above, according to the present invention, edge extraction can be performed satisfactorily without picking up noise other than edge portions.

[Brief explanation of the drawing]

Figure 1 is an overall diagram of the illustration creation device, Figure 2 is a detailed diagram of the image processing section, Figure 3 is a flowchart of the process, and Figure 4 is a C
Display priority order on the RT monitor screen, Fig. 5 shows the differential operator used for processing, Fig. 6 shows the result of region division, Fig. 7 shows an example of the lookup table contents before edge extraction, and Fig. 8 shows after edge extraction. Example of lookup table contents, No. 9
Figure 10 is an explanatory diagram of the lookup table,

Claims (2)

[Claims] (1) An image processing method for extracting edges from given image data, which is characterized by performing tone conversion for edge extraction before edge extraction. (2) Means for supplying image data, tone conversion means for performing tone conversion for edge extraction on the image data supplied by the means, edge extraction of data subjected to tone conversion for edge extraction by the conversion means An image processing device comprising means for performing.