CN103646390B - Image processing system based on multi-level image location and method - Google Patents

Abstract

The invention relates to an image processing system and method based on multi-level image positioning. The system includes an image acquisition device, an image positioning device and an image processing device connected in sequence. The image acquisition device collects a standard sample image and a target image of the same frame, and the image The positioning device sets the feature image and the search range in the standard sample image of the frame according to the grayscale and/or gradient features of each frame of the standard sample image collected, and then performs feature image matching in the target image of the same frame collected Processing, the image processing device performs data processing to obtain the offset of the characteristic image, uses digital image processing technology to compare the collected target image of the same frame with the standard sample image, and then analyzes the comparison result to complete the image processing. The system and method of the invention can perform global positioning, completely align the target image with the standard sample image, and improve the accuracy of image processing and printing plate detection accuracy.

Description

Image processing system and method based on multi-level image positioning

technical field

The invention relates to the field of quality inspection based on machine vision technology, in particular to an image processing system and method based on multi-level image positioning.

Background technique

In order to achieve the long-term goal of free convertibility of RMB, it is an important prerequisite to ensure the quality of its printed products, and the quality of the original plate of banknote glue gravure printing and printing plate is the basis for ensuring the quality of printed products. Any quality problem in any of them will cause damage to the printed products. Quality has serious consequences. The printing plates tested include various offset original plates, offset printing plates, gravure original plates, gravure printing plates, etc. The detection object has a large format, the detection accuracy is extremely high, and the detection speed must be fast.

The way of manual inspection of various printing plates is greatly affected by human subjective factors, the efficiency is low, and missed inspections occur from time to time. With the development of science and technology and the increase of labor costs, the printing plate quality inspection system has gradually appeared to replace it. Manual detection, the existing detection system is to use machine vision technology to compare and analyze the collected target image and standard sample image pixel point, wherein, the collection of target image and standard sample image is usually taken by a camera, although each The shooting should be done in accordance with the agreed position as much as possible, but due to factors such as mechanical errors, electrical movement errors, small angles of the detection objects, vibrations, etc., the target image and the standard sample image taken each time cannot be 100% aligned. If the target image and the standard sample image are not aligned, there will be a deviation in the pixel point comparison between the two, which will affect the subsequent image processing results, and then affect the quality detection accuracy of the printed product.

At present, the existing image positioning method uses a positioning printing device composed of printing plate cylinder, controller and detection head to locate the target image. This positioning method has a complex structure and low positioning accuracy; there are also positioning and image processing through image matching technology. Yes, this positioning method uses a single fixed positioning technology for image matching positioning. Due to the complex and changeable detected objects, some places have dense images, some places have sparse images, some have thicker patterns, and some have thicker patterns. It is thinner, and the single image matching and positioning technology cannot be used to complete accurate positioning for all detection objects, resulting in errors in the image processing results, which cannot meet the requirements for printing plate detection accuracy.

Contents of the invention

Aiming at the problem that the existing detection system cannot accurately locate all detection objects and thus cannot meet the requirements for the detection accuracy of the printing plate, the present invention provides an image processing system based on multi-level image positioning, which can perform global positioning and combine target images with standard The sample image is completely aligned to ensure the accuracy of subsequent image processing and improve the accuracy of printing plate detection. The invention also relates to an image processing method based on multi-level image positioning.

Technical scheme of the present invention is as follows:

An image processing system based on multi-level image positioning, characterized in that it includes an image acquisition device, an image positioning device, and an image processing device connected in sequence, the image acquisition device collects a standard sample image and a target image of the same frame, and the The image positioning device sets the feature image and the search range in the standard sample image of the frame according to the grayscale and/or gradient features of each frame of the standard sample image collected, and then performs the feature image in the target image of the same frame collected. Matching processing, the image processing device performs data processing according to the characteristic image matching result obtained by the image positioning device to obtain the offset of the characteristic image, and adopts digital image processing technology to convert the collected same frame according to the offset of the characteristic image Image comparison is performed between the target image and the standard sample image, and then the comparison result is analyzed to complete the image processing.

The image positioning device performs feature image matching processing based on image distances and/or matching coefficients of grayscale and/or gradients in the target image collected in the same frame.

The image positioning device sets all or most of the target image as a feature image according to the grayscale and/or gradient features of the collected standard sample image, and then performs global matching processing of the feature image in the target image of the same frame collected .

The image positioning device sets the collected standard sample image as a feature image with a sharp gray scale and/or gradient feature, and searches for feature images in a wide range within the same frame of target images collected to perform feature image matching processing.

The image positioning device sets the gray scale and/or image with distinct gradient characteristics of the collected standard sample image as the characteristic image, and there are repeated characteristic images in the target image, and within a small range of the target image collected in the same frame Search feature images for feature image local matching processing.

The image positioning device sets different feature images and search ranges according to the grayscale and/or gradient features of the images in different areas of the collected standard sample images, and performs each feature image respectively in the target image of the same frame collected. The global matching process and/or large-scale search of each feature image to perform feature image matching processing and/or small-scale search of each feature image to perform feature image local matching processing.

The image processing device performs data processing according to the characteristic image matching result obtained by the image positioning device, and performs normalization processing when the characteristic image matching result contains multiple values, according to the image distance and/or matching coefficient of gray scale and/or gradient The weight coefficients are weighted and processed to obtain the offset of the feature image.

An image processing method based on multi-level image positioning, characterized in that the standard sample image and the target image of the same frame are first collected, and according to the grayscale and/or gradient features of each frame of the standard sample image collected in the frame Set the feature image and search range in the standard sample image, and then perform feature image matching processing in the target image of the same frame collected, and then perform data processing according to the obtained feature image matching result to obtain the offset of the feature image, and according to The offset of the characteristic image uses digital image processing technology to perform image comparison between the collected target image and the standard sample image of the same frame, and then analyzes the comparison result to complete the image processing.

Set one or more characteristic images and the search range corresponding to the characteristic images according to the grayscale and/or gradient characteristics of the images in different regions of the collected standard sample images, and in the collected target images of the same frame based on Image distances and/or matching coefficients of grayscale and/or gradients are used for feature image matching processing.

The characteristic image matching process includes the global matching process of the characteristic image and/or searching the characteristic image in a large range to perform the characteristic image matching process and/or searching the characteristic image in a small range to perform the characteristic image local matching process; the global matching process is based on The grayscale and/or gradient features of the standard sample image collected set all or most of the target image as a feature image, and then perform global matching processing of the feature image in the target image of the same frame collected; the large-scale search Feature image for feature image matching processing is to set the grayscale and/or gradient image of the collected standard sample image as the feature image, and search for the feature image in a large range within the target image of the same frame collected to perform the feature image Matching processing; the small-scale search for feature images to perform feature image local matching processing is to set the grayscale and/or gradient characteristic images of the collected standard sample images as feature images, and the target image has repeated features Image, search for feature images in a small range within the target image of the same frame collected to perform feature image local matching processing.

When the feature image matching result contains multiple values, normalization is performed, and the optimal positioning result is selected according to the grayscale and/or gradient image distance and/or matching coefficient, or some or all of the positioning results are integrated as the final positioning result.

Technical effect of the present invention is as follows:

The invention relates to an image processing system based on multi-level image positioning, which is provided with an image acquisition device, an image positioning device and an image processing device connected in sequence, and the image positioning device is based on the grayscale and/or gradient of each frame of standard sample image collected Features Set the feature image and search range in the standard sample image of the frame, and then perform feature image matching processing in the target image of the same frame collected, and the image processing device performs data processing according to the feature image matching result obtained by the image positioning device Get the offset of the feature image, and complete the subsequent image processing. That is, the image positioning device sets the feature image and the search range according to the actual graphic situation of each frame of standard sample image collected, and uses image matching technology to match the feature image, and combines the image processing device to process the matching results to obtain the feature image accurately offset, so that the target image can be completely aligned with the standard sample image, which provides a basis for subsequent image processing. The grayscale and/or gradient features of the standard sample image can dynamically and flexibly set the feature image and search range to achieve multiple images. Hierarchical global positioning ensures the accuracy of image processing, further improves the detection accuracy of printing plates, saves the complicated positioning printing device used in the existing positioning method, and more importantly, solves the problem that the existing detection system cannot target The precise positioning of all detection objects cannot meet the requirements for the detection accuracy of printing plates. Even when the image of the detection object is dense and uneven, and the thickness of the pattern is uneven and complex, the system of the present invention can directly use the image positioning device according to the grayscale and/or or gradient features manually or automatically set the feature image and search range in the standard sample image of the frame, and use image matching technology to process data to obtain the offset of the feature image, realize the target image positioning, and have high reliability. The quality of the gravure original plate and printing plate will lay a solid foundation for realizing the digitization and standardization of banknote design original plate and printing plate production.

The present invention also relates to an image processing method based on multi-level image positioning, in which a feature image and a search range are set in the standard sample image of the frame according to the grayscale and/or gradient features of each frame of standard sample image collected, Furthermore, the characteristic image matching process is carried out in the target image of the same frame collected, and then data processing is performed according to the obtained characteristic image matching result to obtain the offset of the characteristic image, so that the positioning of the target image and the standard sample image can be realized, that is, the The multi-level global positioning of the detection object makes the target image and the standard sample image completely aligned, which is suitable for the situation where the detection object image is flexible and changeable, and then digital image processing technology is used to image the same frame of the target image and the standard sample image comparison, and then analyze the comparison results, which improves the accuracy of image processing and further improves the accuracy of printing plate quality detection.

Description of drawings

FIG. 1 is a schematic structural diagram of an image processing system based on multi-level image positioning in the present invention.

Fig. 2a and Fig. 2b are respectively the working principle diagrams of the first kind of positioning of the standard sample image and the target image of the image processing system based on the multi-level image positioning of the present invention.

Fig. 3a and Fig. 3b are respectively the working principle diagrams of the second positioning of the standard sample image and the target image of the image processing system based on multi-level image positioning in the present invention.

Fig. 4a and Fig. 4b are diagrams of the third positioning working principle of the standard sample image of the image processing system based on multi-level image positioning according to the present invention.

Fig. 5a and Fig. 5b are diagrams showing the working principles of the third positioning of the target image of the image processing system based on multi-level image positioning according to the present invention.

Fig. 6 is a working flowchart of the image processing method based on multi-level image positioning in the present invention.

detailed description

The present invention will be described below in conjunction with the accompanying drawings.

The invention relates to an image processing system based on multi-level image positioning, which is used for image processing when performing quality inspection on printing plates. The structural diagram of the system is shown in Figure 1, which includes an image acquisition device, an image positioning device, and an image processing device connected in sequence. The drive of the electrical device moves downwards within the full-page range, and collects high-quality images during the movement or at a fixed point. The image acquisition device collects the standard sample image and the target image of the same frame, and the image positioning device according to the grayscale of each frame of the standard sample image collected Set the feature image and search range in the standard sample image of the frame, and then perform feature image matching processing in the target image of the same frame collected; the image processing device obtains the feature image according to the image positioning device Perform data processing on the matching result to obtain the offset of the characteristic image, and use digital image processing technology to compare the target image of the same frame collected with the standard sample image (such as image alignment, Compare the pixel values point by point, of course, you can also use smoothing and blurring, etc.), and then analyze the comparison results to complete the image processing.

The system of the present invention adopts multi-level global positioning technology. After the image acquisition device collects a frame of target image, it first locates the target image through the image positioning device. The positioning result of the frame target image is used for image processing such as image template learning or pattern detection in the next step. Specifically, the image positioning device uses image matching processing technology to use three working methods:

(1) The first method is global matching, that is, the image positioning device sets all or most of the target image as the feature image according to the grayscale and/or gradient characteristics of the collected standard sample image, and sets the target image to be positioned and All or most of the standard sample images are compared, and then the global matching process of the feature image is performed based on the grayscale and/or gradient image distance and/or matching coefficient in the target image of the same frame collected. The image processing device performs data processing according to the characteristic image matching result obtained by the image positioning device, calculates the image distance and/or correlation matching coefficient based on color and/or gradient, and obtains the offset of the best point of the matching result, or combines multiple methods The calculated offset is used to obtain the offset of the feature image, which can be used as the horizontal and vertical offsets of the two images.

As shown in Fig. 2a and Fig. 2b for example, Fig. 2a is a standard sample image, and a large-area feature image is inside a thick line frame, that is, the image positioning device sets most of the standard sample image as a feature image; Fig. 2b is a feature image to be The detected and localized target image, the thin line box is the matching result.

(2) The second method uses a positioning kernel (also known as a feature image) for matching, that is, the image positioning device sets the grayscale and/or gradient image of the collected standard sample image as a feature image, and the positioning kernel can It is automatically calculated or manually specified, and the positioning kernel is searched in a large range in the target image to be located, and then the feature image is matched based on the grayscale and/or gradient image distance and/or matching coefficient in the target image of the same frame collected deal with. The image processing device performs data processing according to the characteristic image matching result obtained by the image positioning device, calculates the image distance and/or correlation matching coefficient based on color and/or gradient, and obtains the offset of the best point of the matching result, or combines multiple methods The calculated offset is used to obtain the horizontal and vertical offsets of the localization core. There can be multiple positioning kernels, and the results of multiple positioning kernels can be optimized or integrated.

The specific embodiment is shown in Figure 3a and Figure 3b, Figure 3a is a standard sample image, the positioning kernel (feature image) is inside the thick line frame, and the thin line frame surrounding the thick line frame is the search range; Figure 3b is the image to be detected and positioned The target image, the thin line frame is the matching result.

(3) The third method uses a local positioning kernel (also known as a local feature image) for matching, that is, the image positioning device sets the grayscale and/or gradient characteristic image of the collected standard sample image as the feature image, and the positioning The kernel can be automatically calculated or manually specified, and the positioning kernel is searched in a small range in the target image to be located, and then the feature image is based on the image distance and/or matching coefficient of the grayscale and/or gradient in the target image collected in the same frame. Partial matching processing. The image processing device performs data processing according to the characteristic image matching result obtained by the image positioning device, calculates the image distance and/or correlation matching coefficient based on color and/or gradient, and obtains the offset of the best point of the matching result, or combines multiple methods The calculated offset is used to obtain the horizontal and vertical offsets of the local positioning kernel. There can be multiple local positioning kernels, and the results of multiple localization kernels can be optimized or integrated. The difference between the third method and the second method is that the former has a small search range and is suitable for the positioning of repeated feature images. It needs to rely on the positioning results of adjacent images (using the favorable fact that the offsets of adjacent frame images are basically the same), Or use the preliminary positioning results obtained by other technical methods to perform fine positioning.

An example of a localization result by means of adjacent images is shown in FIGS. 4a, 4b, 5a and 5b. Among them, Fig. 4a and Fig. 4b are two frames of standard sample images collected and photographed at adjacent positions, and Figs. 5a and 5b are two frames of target images to be detected collected and photographed at adjacent positions, corresponding to Figs. 4a and 4b respectively. The thick line frame in Figure 4b is the feature image, and the thin line frame surrounding the thick line frame is the search range. The frame image is positioned using the second method (localization kernel matching), and the target image of the frame corresponding to Figure 4b is shown in the figure 5b. The thick-lined frame in Figure 4a is the local positioning nucleus, and the thin-lined frame surrounding the thick-lined frame is the search range. It can be seen that the search range is very small, because the positioning nucleus pattern is repeated nearby, and the target in the frame corresponding to Figure 4a The image is shown in Figure 5a. In the positioning detection, the positioning kernel matching can be performed on Figure 5b first, and the matching result is shown in the thin line box in Figure 5b; Search range, the search range in Figure 5a can be obtained as shown in the larger line frame in the figure, and the smaller line frame in Figure 5a is the matching result.

For a certain feature image used by any of the above positioning techniques, after the image positioning device locates, the image processing device performs data processing according to the matching result of the feature image obtained by the image positioning device. values, such as color distance, color correlation coefficient, gradient distance, gradient correlation matching coefficient, etc., they are normalized, and then weighted according to the weight coefficients such as grayscale and/or gradient image distance and/or matching coefficient, etc. and, as the matching result.

The image positioning device sets different feature images and search ranges according to the grayscale and/or gradient features of the images in different areas of the collected standard sample images, and performs global positioning of each feature image in the target image of the same frame collected. Matching processing and/or searching each feature image in a large range to perform feature image matching processing and/or searching each feature image in a small range to perform feature image local matching processing. The system of the present invention can configure the corresponding positioning method according to the grayscale and/or gradient characteristics of each frame image (standard sample image and target image), or select the image according to the size of each pixel value in each frame image The working mode of the positioning device and the image processing device can use one of the above-mentioned working states alone, or use two or three of them in combination, and synthesize and select their calculation results.

The image acquisition device of the system preferably includes an area array camera and a laser displacement sensor. The area array camera can be a CCD camera or a CMOS camera. Contrast measurement of the pattern, pattern width measurement and measurement of the amount of miss. Through the guide rail setting, the area array camera and the laser displacement sensor move in the horizontal plane parallel to the detection platform above the detection platform. The area array camera automatically scans the detection object placed flat on the detection platform for pattern detection to obtain high-quality collected images. An array camera or a laser displacement sensor collects the depth of the detected object for depth measurement.

The present invention also relates to an image processing method based on multi-level image positioning. This method corresponds to the above-mentioned image processing system based on multi-level image positioning. Its flow chart is shown in FIG. The target image, and set the feature image and search range in the standard sample image of the frame according to the grayscale and/or gradient features of each frame of the standard sample image collected, and then perform feature extraction in the target image of the same frame collected Image matching processing, and then perform data processing according to the obtained feature image matching results to obtain the offset of the feature image, and use digital image processing technology to convert the target image and standard sample image of the same frame collected according to the offset of the feature image Perform image comparison, and then analyze the comparison results to complete image processing.

The method of the present invention can use the area array camera and the laser displacement sensor to carry out the image acquisition of the target image and the standard sample object, and set the guide rail to make the area array camera and the laser displacement sensor move in the horizontal plane and the vertical direction, and the area array camera automatically scans The detection object placed flat on the detection platform can collect high-quality images in the horizontal plane for pattern detection, and the area array camera or laser displacement sensor can collect the depth of the detection object in the vertical direction for depth measurement, which can use high-resolution Microscopic technology and computer image processing technology, through comparison with standard samples, can detect the pattern quality, layout size and layout defects of various original plates and printing plates.

The method preferably sets one or more characteristic images and a search range corresponding to the characteristic images according to the gray scale and/or gradient characteristics of the image in different regions of the collected standard sample image, and the target in the same frame collected The characteristic image matching processing is performed based on the image distance and/or matching coefficient of grayscale and/or gradient in the image. The feature image matching process mentioned here includes one of the three processing methods of global matching processing of feature images, large-scale search of feature images for feature image matching processing, and small-scale search of feature images for feature image local matching processing. Two or three. The global matching process is to set all or most of the target image as a feature image according to the grayscale and/or gradient features of the collected standard sample image, and then perform global matching of the feature image in the target image of the same frame collected Processing; the large-scale search for feature images to carry out feature image matching processing is to set the grayscale and/or gradient characteristic images of the collected standard sample images as feature images, and to collect large-scale images within the same frame of the target image Searching for feature images to perform feature image matching processing; the small-scale search for feature images to perform local feature image matching processing is to set the collected standard sample images with grayscale and/or images with distinct gradient features as feature images, and the There are repeated feature images in the target image, and the feature images are searched in a small range in the target image of the same frame collected to perform local matching processing of feature images.

The method of the present invention can select one or more positioning and image processing methods according to the actual conditions of each frame of image, and perform normalization processing when a certain feature image matching result contains multiple values, according to grayscale and/or gradient The image distance and/or the weight coefficient of the matching coefficient are weighted and processed to obtain the offset of the feature image.

It should be pointed out that the specific embodiments described above can enable those skilled in the art to understand the invention more comprehensively, but do not limit the invention in any way. Therefore, although this specification has described the invention in detail with reference to the accompanying drawings and embodiments, those skilled in the art should understand that the invention can still be modified or equivalently replaced. The technical solutions and their improvements in the spirit and scope should all be included in the protection scope of the invention patent.

Claims (9)

1. an image processing system based on multi-level image location, it is characterised in that include the image collector being sequentially connected with Put, image positioning device and image processing apparatus, the master sample image of described image acquisition device same number of frames and target figure Picture, described image positioning device according to the gray scale of each frame master sample image gathered and/or Gradient Features in the standard of described frame Sample image sets characteristic image and hunting zone, and then carries out characteristic image coupling in the target image of the same number of frames gathered Processing, described image processing apparatus carries out data process according to the characteristic image matching result that image positioning device obtains and obtains feature The side-play amount of image, and use digital image processing techniques by the target of the same number of frames of collection according to the side-play amount of described characteristic image Image and master sample image carry out image comparison, then comparison result has been analyzed image procossing；

Described image positioning device in the zones of different of the master sample image gathered according to the gray scale of image and/or Gradient Features Set different characteristic images and hunting zone, and carry out the complete of each characteristic image respectively in the target image of the same number of frames gathered Office's matching treatment and/or each characteristic image of extensive search are to carry out characteristic image matching treatment and/or each characteristic pattern of little range searching As to carry out characteristic image local matching process.

System the most according to claim 1, it is characterised in that described image positioning device is in the target of the same number of frames gathered Characteristic image matching treatment is carried out in image based on gray scale and/or the image distance of gradient and/or matching factor.

System the most according to claim 1 and 2, it is characterised in that described image positioning device is according to the standard sample gathered Target image is wholly or largely set as characteristic image by the gray scale of this image and/or Gradient Features, and then identical gather The global registration carrying out characteristic image in the target image of frame processes.

System the most according to claim 1 and 2, it is characterised in that the master sample that described image positioning device will gather The image setting of the gray scale of image and/or Gradient Features distinctness is characterized image, in the target image of the same number of frames gathered on a large scale Search characteristics image is to carry out characteristic image matching treatment.

System the most according to claim 1 and 2, it is characterised in that the master sample that described image positioning device will gather The image setting of the gray scale of image and/or Gradient Features distinctness is characterized image, has the characteristic pattern of repetition in described target image Picture, in the target image of the same number of frames gathered, little range searching characteristic image is to carry out characteristic image local matching process.

System the most according to claim 1, it is characterised in that described image processing apparatus obtains according to image positioning device Characteristic image matching result carry out data process, and carry out standardization processing when characteristic image matching result comprises multiple value, It is weighted according to the weight coefficient of gray scale and/or the image distance of gradient and/or matching factor and processes obtaining the inclined of characteristic image Shifting amount.

7. an image processing method based on multi-level image location, it is characterised in that first gather the master sample figure of same number of frames Picture and target image, and according to the gray scale of each frame master sample image gathered and/or Gradient Features at the master sample of described frame Image sets characteristic image and hunting zone, and then carries out characteristic image matching treatment in the target image of the same number of frames gathered, Then carry out data process according to the characteristic image matching result obtained and obtain the side-play amount of characteristic image, and according to described characteristic pattern The side-play amount of picture uses digital image processing techniques that target image and the master sample image of the same number of frames of collection are carried out image ratio Right, then comparison result has been analyzed image procossing；

In the zones of different of the master sample image gathered, gray scale and/or Gradient Features according to image set one or more spies Levy image and the hunting zone corresponding with described characteristic image, and in the target image of the same number of frames gathered based on gray scale and/or The image distance of gradient and/or matching factor carry out characteristic image matching treatment.

Method the most according to claim 7, it is characterised in that described characteristic image matching treatment includes the complete of characteristic image Office matching treatment and/or extensive search characteristic image with carry out characteristic image matching treatment and/or little range searching characteristic image with Carry out characteristic image local matching process；The process of described global registration is gray scale and/or the gradient of the master sample image according to gathering Feature is wholly or largely set as characteristic image by target image, and then carries out spy in the target image of the same number of frames gathered The global registration levying image processes；Described extensive search characteristic image is the standard that will gather to carry out characteristic image matching treatment The image setting of the gray scale of sample image and/or Gradient Features distinctness is characterized image, big in the target image of the same number of frames gathered Range searching characteristic image is to carry out characteristic image matching treatment；Described little range searching characteristic image is to carry out characteristic image local Matching treatment is that the gray scale of the master sample image of collection and/or the image setting of Gradient Features distinctness are characterized image, described mesh Having the characteristic image of repetition in logo image, in the target image of the same number of frames gathered, little range searching characteristic image is to carry out spy Levy image local matching treatment.

Method the most according to claim 8, it is characterised in that advise when characteristic image matching result comprises multiple value Generalized processes, and chooses oplimal Location result, or comprehensive part or complete according to gray scale and/or the image distance of gradient and/or matching factor Portion's positioning result is as final positioning result.