WO2019057067A1 - 图像质量评估方法及装置 - Google Patents

图像质量评估方法及装置 Download PDF

Info

Publication number
WO2019057067A1
WO2019057067A1 PCT/CN2018/106451 CN2018106451W WO2019057067A1 WO 2019057067 A1 WO2019057067 A1 WO 2019057067A1 CN 2018106451 W CN2018106451 W CN 2018106451W WO 2019057067 A1 WO2019057067 A1 WO 2019057067A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
quality evaluation
target area
quality assessment
evaluated
Prior art date
Application number
PCT/CN2018/106451
Other languages
English (en)
French (fr)
Inventor
姜兴
李宏宇
朱帆
Original Assignee
众安信息技术服务有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 众安信息技术服务有限公司 filed Critical 众安信息技术服务有限公司
Priority to SG11201907815V priority Critical patent/SG11201907815VA/en
Priority to JP2020504760A priority patent/JP2020513133A/ja
Publication of WO2019057067A1 publication Critical patent/WO2019057067A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/20Image enhancement or restoration using local operators
    • G06T5/30Erosion or dilatation, e.g. thinning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/25Determination of region of interest [ROI] or a volume of interest [VOI]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/20Image preprocessing
    • G06V10/28Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/40Extraction of image or video features
    • G06V10/44Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10004Still image; Photographic image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20092Interactive image processing based on input by user
    • G06T2207/20104Interactive definition of region of interest [ROI]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30168Image quality inspection

Definitions

  • the present invention relates to the field of digital image processing technologies, and in particular, to an image quality evaluation method and apparatus.
  • the current image quality assessment method is basically based on processing the entire image to obtain its corresponding quality index.
  • the quality assessment of computers in this way does not really reflect the process of human vision.
  • the human eye vision system implements autofocus when evaluating surrounding scenes. This focusing process is equivalent to finding the target area of real interest, that is, target detection.
  • the quality of the region of interest (ROI) in the image is more meaningful for practical applications.
  • the present invention is directed to the above problems, and proposes an image quality evaluation method and apparatus.
  • An aspect of the present invention provides an image quality evaluation method, the method comprising: performing quality assessment separately on at least one target region in an image to be evaluated to determine a quality assessment result of the at least one target region; based on the determined Determining a quality assessment result of the at least one target area, and performing quality assessment on the image to be evaluated.
  • the method further comprises: performing target detection on the image to be evaluated to determine the at least one target area.
  • the method for performing target detection on the image to be evaluated includes any one of the following methods: based on a traditional feature extraction method and a depth feature extraction method.
  • the step of performing target detection on the image to be evaluated comprises: classifying a contour of the binarized image obtained from the image to be evaluated to determine a text outline; based on the determined text contour, The text is merged in a specified direction to determine the at least one target area.
  • the contour of the binarized image is classified based on parameters determined by at least one of: determining a number of non-zero pixels within a contour of the binarized image; determining the two And an aspect ratio and an aspect ratio of the contour of the image; determining a number of similar width contours and a number of similar height contours present within a specified direction neighborhood within a specified direction range of the contour of the binarized image.
  • the step of merging the text in the specified direction based on the determined text outline comprises: expanding the determined text outline in a specified direction by setting an expansion operator and a corrosion operator and Corrosion operation.
  • the step of separately performing quality assessment on at least one target region in the image to be evaluated comprises: determining a quality assessment of the at least one target region based on the determined at least one target region and an image quality assessment model result.
  • the step of determining a quality assessment result of the at least one target region based on the determined at least one target region and image quality assessment model comprises: determining an image quality assessment model using a depth learning based training method And determining a quality assessment result of the at least one target area based on the determined at least one target area and the image quality assessment model.
  • the step of separately performing quality assessment on the at least one target region comprises performing quality assessment on the at least one target region based on statistics on pixel grayscale values.
  • the step of performing quality assessment on the image to be evaluated based on the determined quality assessment result of the at least one target region comprises: based on the determined at least one by mass weighted averaging The quality assessment result of the target area is used to perform quality assessment on the image to be evaluated.
  • the step of separately performing quality assessment on the at least one target area to determine a quality assessment result of the at least one target area comprises: using a non-reference quality evaluation indicator, respectively, for the at least one target area Performing a quality assessment to determine a quality assessment result of the at least one target area, wherein the no-reference quality evaluation indicator includes at least one of an edge intensity, a noise rate, or a uniform brightness distribution.
  • Another aspect of the present invention provides an image quality evaluation apparatus, the apparatus comprising: a target area quality evaluation unit configured to perform quality evaluation separately on at least one target area in an image to be evaluated to determine the at least one target area a quality evaluation result; the image quality evaluation unit to be evaluated is configured to perform quality evaluation on the image to be evaluated based on the determined quality evaluation result of the at least one target area.
  • the apparatus further includes: a target detecting unit configured to perform target detection on the image to be evaluated to determine the at least one target area.
  • the method for performing target detection on the image to be evaluated by the target detecting unit includes any one of the following methods: based on a traditional feature extraction method and a depth feature extraction method.
  • the target detecting unit includes: a contour classifying unit configured to classify an outline of the binarized image obtained from the image to be evaluated to determine a text outline; and a text merging unit configured to The text is merged in a specified direction based on the determined text outline to determine the at least one target area.
  • the contour classification unit is further configured to: classify a contour of the binarized image based on parameters determined by at least one of: determining a contour within the contour of the binarized image a non-zero pixel number; determining an aspect ratio and an aspect ratio of the contour of the binarized image; and determining a similar width existing within a specified direction neighborhood within a specified direction range of the contour of the binarized image The number of contours and the number of similar height contours.
  • the text merging unit is further configured to perform an expansion operation and a etch operation on the determined text contour in a specified direction by setting an expansion operator and a corrosion operator.
  • the target area quality evaluation unit is further configured to determine a quality assessment result of the at least one target area based on the determined at least one target area and an image quality assessment model.
  • the target area quality evaluation unit is further configured to: determine an image quality evaluation model by using a depth learning based training method, and based on the determined at least one target area and the image quality evaluation model Determining a quality assessment result of the at least one target area.
  • the target area quality evaluation unit is further configured to perform quality assessment on the at least one target area based on statistics on pixel gray values.
  • the image quality evaluation unit to be evaluated is further configured to: perform quality on the image to be evaluated based on the determined quality evaluation result of the at least one target region by means of quality weighted averaging Evaluation.
  • the target area quality evaluation unit is further configured to perform quality assessment on the at least one target area separately using a non-reference quality evaluation indicator to determine a quality assessment result of the at least one target area.
  • the non-reference quality evaluation indicator includes at least one of an edge intensity, a noise rate, or a uniform brightness distribution.
  • Another aspect of the present invention also provides a computer device including a memory, a processor, and a computer program stored on the memory by the processor, the processor executing the computer program to implement any of the above The method described in the item.
  • Another aspect of the present invention also provides a computer storage medium having stored thereon a processor executable program that, when executed by the processor, implements the method of any of the above.
  • the image quality evaluation method provided by the invention makes the image quality evaluation process focus on the target region of interest and ignore the image quality of the insignificant region, thereby realizing the quality evaluation of the image to be evaluated, the evaluation speed is fast, the evaluation accuracy is high, and the objective The quality of the image to be evaluated is effectively evaluated so that the quality assessment results are as consistent as possible with the perception of the human eye.
  • FIG. 1 is a flow chart of an image quality evaluation method according to an embodiment of the present invention.
  • FIG. 2 is a flow chart of another method of image quality assessment in accordance with an embodiment of the present invention.
  • FIG. 3 is a flow chart of a text image quality assessment method in accordance with an embodiment of the present invention.
  • FIG. 4 is an illustration of a text image in accordance with an embodiment of the present invention.
  • Figure 5 is a binarized image of the text image of Figure 4.
  • FIG. 6 is a schematic diagram of a text line area obtained after performing target detection in FIG. 4.
  • FIG. 6 is a schematic diagram of a text line area obtained after performing target detection in FIG. 4.
  • FIG. 7 is a schematic diagram of a composite image of an image quality assessment model in accordance with an embodiment of the present invention.
  • FIG. 8 is an evaluation result of the text image of FIG. 4.
  • FIG 9 is an illustration of another text image in accordance with an embodiment of the present invention.
  • Figure 10 is a binarized image of the text image of Figure 9.
  • FIG. 11 is a schematic diagram of a text line area obtained after performing target detection in FIG. 9.
  • FIG. 11 is a schematic diagram of a text line area obtained after performing target detection in FIG. 9.
  • Figure 12 is a result of evaluation of the text image of Figure 9.
  • Figure 13 is a schematic diagram of an image quality evaluation apparatus according to an embodiment of the present invention.
  • FIG. 1 is a flow chart of an image quality evaluation method according to an embodiment of the present invention.
  • S101 Perform quality assessment on at least one target area in the image to be evaluated to determine a quality assessment result of each target area in the at least one target area;
  • S102 Perform quality assessment on the image to be evaluated based on the determined quality assessment result of each target region of the at least one target region.
  • the manner of determining the target area may be preset, or may be performed before each evaluation.
  • FIG. 2 is a flow chart of another method of image quality assessment in accordance with an embodiment of the present invention.
  • the present invention provides a quality assessment method that includes the steps shown in Figure 1:
  • S201 Perform target detection on the image to be evaluated to determine at least one target area
  • S202 Perform quality assessment on the at least one target area separately to determine a quality assessment result of each target area in the at least one target area;
  • S203 Perform quality assessment on the image to be evaluated based on the determined quality assessment result of each target region of the at least one target region.
  • the image to be evaluated may be subjected to target detection based on traditional feature extraction method, target detection based on depth feature extraction method (such as CTPN method, Haar classifier, Fater-rcnn), etc., wherein the image to be evaluated may include any object, for example, Animals, faces, food, cars or text lines, etc.
  • the image to be evaluated herein may be a color image or a grayscale image.
  • the result returned after the target detection is a plurality of target regions of interest, which are saved as images.
  • the quality assessment uses non-reference quality evaluation indicators, including edge strength, noise rate or uniform brightness distribution.
  • a machine learning-based approach can be used to train the quality assessment model prior to quality assessment for quality assessment of the input image to be evaluated.
  • Model training needs to collect various target area images in advance, and perform quality labeling.
  • the size of the mass labeling is represented by an integer ranging from 0 to 100. The larger the labeling value, the better the corresponding image quality. It should be understood that it can also be used. Other suitable ways to define the size of the quality annotation.
  • model training employs a deep learning (Convolutional Neural Network CNN) based training method to determine a higher accuracy image quality model.
  • weighted average is performed on the quality assessment result of each target quality region in the at least one target region.
  • the method can respectively assign weights according to the importance of each target area, and the importance of the target area depends on the size and definition of the target area, the degree of interest in the target area, and the like.
  • the weighted average manner evolves into the arithmetic mean of the quality assessment results for the respective target regions.
  • FIG. 3 is a flow chart of a text image quality evaluation method according to the present invention. As shown in FIG. 2, the method includes the following steps:
  • the image to be evaluated is a text image (for example, as shown in FIG. 4 or FIG. 9), the text line is an object of interest in the image, and the purpose of target detection of the text line is to detect the position of the text line in the image. .
  • step S303 If the image to be evaluated is a color image, step S303 is performed, otherwise skip S303 to perform step S304;
  • step S302 when the input image to be evaluated is a grayscale image (for example, as shown in FIG. 4 and FIG. 9), the image to be evaluated is the grayscale image to be processed, and after step S302 is performed, the process proceeds directly to step S303.
  • step S304 the image to be evaluated is binarized; when the image to be evaluated is a color image, the image to be evaluated is an image subjected to gradation processing in step S303.
  • the local adaptive binarization method is used to perform binarization processing on the grayscale image, that is, the binarization threshold on the pixel position is determined according to the pixel value distribution of the pixel neighborhood, and the grayscale image to be processed is processed. For all pixels, do the following:
  • N is between [2, 5];
  • the pixel value is compared to the threshold obtained in the previous step. If the pixel value is greater than the threshold, the pixel value is set to 255 in the image, otherwise the pixel value is set to 0 in the image. After binarizing the images of FIGS. 4 and 9, the corresponding binarized images as shown in FIGS. 5 and 10 can be obtained.
  • S305 classify all the contours of the binarized image and remove the non-text type contour to determine the outline of the text, and perform the following operations on each contour in the binarized image:
  • the contour is considered to be a non-text contour, at a binary value Set the pixel value of all positions in the outline to 0 in the image.
  • the first threshold is between [2, 5]
  • the second threshold is between [8, 12]
  • the third threshold is between [2, 5]
  • the fourth threshold is [8, Between 12]
  • the fifth threshold is between [2, 5].
  • a similar width profile of a profile having a width W and a height H is defined as a profile width ranging between [0.7 W, 1.3 W] and a profile height of a similar height profile being between [0.7H, 1.3H].
  • the contour in the binarized image processed in step S305 is subjected to an expansion operation and a etching operation in a horizontal direction (it should be understood that the vertical direction or any suitable direction may also be performed), Get the text line area.
  • the expansion operation is to add a boundary to the operation object
  • the erosion is to delete some pixels of the object boundary, wherein the definition of the boundary is given by the corresponding operation operator, for example, if the size of the expansion operator is 5 ⁇ 1, The pixel is centered, and pixels in the range of 5 ⁇ 1 are set as target pixels.
  • steps S304 and S305 are completed, 20 expansion operations are performed by using an expansion operator with a size of 5 ⁇ 1, and the size is The 5 ⁇ 1 corrosion operator performs 15 etching operations, and the circumscribed rectangle of all the contours is used as a mask to obtain the text line area of the image. It should be understood that in the specific operation, those skilled in the art can expand the expansion operator and expand the expansion operator. The number of operations and the number of corrosions are appropriately adjusted. 6 and 11 are text line areas obtained after the image to be evaluated is completed in step S306.
  • S307 Perform quality assessment on the text line area detected in S306.
  • an image quality assessment model oriented to the text line region is pre-trained using a deep learning based approach.
  • the training text line data can be synthesized by itself, or the annotation can be directly intercepted from the text image.
  • the self-synthesized text line first randomly selects the candidate characters from the commonly used Chinese words, English words and common punctuation marks in Chinese and English, and then combines the strings with different background images, adds different degrees of blur, and then compresses them into Different levels of quality and preservation.
  • the quality parameter of the composite image the quality of the corresponding image is marked, and the size is between 0 and 100.
  • FIG. 7 is a schematic diagram of the composite image.
  • the 10 text line regions in FIG. 6 and the 4 text line regions in FIG. 11 are respectively input into the trained image quality evaluation model of the text-oriented line region, and the corresponding text line region quality evaluation result is obtained.
  • 8 shows the quality evaluation results of all the text line areas line_1, line_2, line_3, line_4, line_5, line_6, line_7, line_8, line_9, and line_10 of FIG. 6, and
  • FIG. 12 shows all the text line areas line_1a, line_2a of FIG.
  • the quality assessment results of line_3a and line_4a are respectively input into the trained image quality evaluation model of the text-oriented line region, and the corresponding text line region quality evaluation result is obtained.
  • 8 shows the quality evaluation results of all the text line areas line_1, line_2, line_3, line_4, line_5, line_6, line_7, line_8, line_9, and line_10 of FIG. 6, and
  • FIG. 12 shows all the text line areas line_1a, line_2a of FIG.
  • S308 Perform quality assessment on the image to be evaluated by means of quality weighted average based on the quality assessment result of the text line region obtained in S307;
  • the weighting value is determined according to the importance of the target area.
  • the importance of the target area is assumed to be the same, so that the weighted average mode is simplified to the arithmetic average mode, and the quality evaluation result of the text image of FIG. 4 is a figure.
  • the average value of the quality evaluation results of all the text line areas line_1, line_2, line_3, line_4, line_5, line_6, line_7, line_8, line_9 and line_10 shown in 7 is 16 and the quality evaluation result of the text image of Fig. 9 is the figure.
  • the average value of the quality evaluation results of all the text line areas line_1a, line_2a, line_3a, and line_4a shown in 12 is 97.
  • the quality evaluation method of the above embodiment is adopted, and the target area of interest is paid attention to, the evaluation speed is fast, the evaluation accuracy is high, the quality of the text image is effectively evaluated, and the text image with different quality is conveniently processed. .
  • the flow of quality assessment may also include steps S201, S202, and S203.
  • a depth learning based algorithm may be employed in the target detection step.
  • the quality of the target area can be evaluated based on the statistics of the gray value of the pixel, for example, using the Laplacian variance algorithm.
  • the quality evaluation result based on the target region may be used, and the quality of the image to be evaluated is evaluated by mass weighted average.
  • Figure 13 is a schematic diagram of an image quality evaluation apparatus according to an embodiment of the present invention.
  • the present invention also provides a quality evaluation device 1200 as shown in FIG. 13, which includes a target detection unit 1201, a target region quality evaluation unit 1202, and an image quality evaluation unit 1203 to be evaluated.
  • the target detecting unit 1201 is configured to detect an image to be evaluated to confirm the target area.
  • the target area quality evaluating unit 1202 is configured to perform quality evaluation on the target area determined in the target detecting unit 1201 to determine a quality evaluation result of the target area.
  • the image quality evaluation unit to be evaluated 1203 is configured to perform quality evaluation on the image to be evaluated based on the quality evaluation result of the target region determined in the target region quality evaluation unit 1202.
  • the image quality evaluation unit 1203 to be evaluated may perform quality evaluation on the image to be evaluated by mass weighted average.
  • the target detecting unit 1201 includes a contour classifying module 1202a and a text combining module 1202b.
  • the contour classification unit 1202a is configured to classify the contour of the binarized image based on parameters determined by at least one of the following operations to determine a text contour: determining a non-zero pixel number nonz within the contour of the binarized image; Determining an aspect ratio hw and an aspect ratio wh of the contour of the binarized image; and determining the number of similar width profiles and similar height profiles present within a specified direction neighborhood within a specified direction range of the contour of the binarized image The number SH.
  • the text merging unit 1202b is further configured to perform an expansion operation and a etch operation on the determined text contour in a specified direction by setting an expansion operator and a corrosion operator.
  • the text merging unit 1202b is configured to merge the text in a specified direction based on the text outline determined in the contour categorizing unit 1202a to determine the target area.
  • the target area quality assessment unit 1202 is further configured to determine a quality assessment result of the target area based on the determined target area and the image quality assessment model.
  • the target area quality assessment unit 1202 is further configured to perform quality assessment of the target area based on statistics of pixel grayscale values.
  • the flow of the quality assessment method of Figures 1, 2, 3 also represents machine readable instructions comprising a program executed by a processor.
  • the program can be embodied in software stored on a tangible computer readable medium such as a CD-ROM, floppy disk, hard disk, digital versatile disk (DVD), Blu-ray disk or other form of memory.
  • a tangible computer readable medium such as a CD-ROM, floppy disk, hard disk, digital versatile disk (DVD), Blu-ray disk or other form of memory.
  • some or all of the example methods in Figures 1, 2 may utilize an application specific integrated circuit (ASIC), programmable logic device (PLD), field programmable logic device (EPLD), discrete logic, hardware, firmware Any combination of the like is implemented.
  • ASIC application specific integrated circuit
  • PLD programmable logic device
  • EPLD field programmable logic device
  • the example processes of Figures 1, 2 can be implemented using coded instructions (such as computer readable instructions) stored on a tangible computer readable medium, such as a hard disk, flash memory, read only memory (ROM), optical disk. (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage medium on which information can be stored for any time (eg, long, permanent, transient) Situation, temporary buffering, and/or caching of information).
  • a tangible computer readable medium such as a hard disk, flash memory, read only memory (ROM), optical disk. (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage medium on which information can be stored for any time (eg, long, permanent, transient) Situation, temporary buffering, and/or caching of information).
  • a tangible computer readable medium such as a hard disk, flash memory, read only memory (ROM), optical disk. (CD), digital versatile disc (DVD), cache, random access memory (RAM), and/or any other storage medium on which information
  • Figures 1, 2 may be implemented with encoded instructions (such as computer readable instructions) stored on a non-transitory computer readable medium, such as a hard disk, flash memory, read only memory, optical disk, Digital versatile disc, cache, random access memory and/or any other storage medium in which information can be stored at any time (eg, long time, permanently, transient, temporary buffering, and/or caching of information). ).
  • a non-transitory computer readable medium such as a hard disk, flash memory, read only memory, optical disk, Digital versatile disc, cache, random access memory and/or any other storage medium in which information can be stored at any time (eg, long time, permanently, transient, temporary buffering, and/or caching of information).

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)
  • Testing, Inspecting, Measuring Of Stereoscopic Televisions And Televisions (AREA)

Abstract

一种图像质量评估方法及装置,该方法包括:对待评估图像进行目标检测,以确定至少一个目标区域;对该至少一个目标区域分别进行质量评估,以确定该至少一个目标区域的质量评估结果;基于所确定的该至少一个目标区域的质量评估结果,对待评估图像进行质量评估。该方法使得图像质量评估过程关注感兴趣的目标区域,忽略无关紧要的区域的图像质量,从而实现对待评估图像进行质量评估,评估速度快,评估精度较高,客观有效地评估了待评估图像的质量。

Description

图像质量评估方法及装置
本申请要求2017年09月20日提交的申请号为No.201710854415.9的中国申请的优先权,通过引用将其全部内容并入本文。
技术领域
本发明涉及数字图像处理技术领域,具体涉及一种一种图像质量评估方法及装置。
发明背景
随着数码相机、监控摄像头、手机等数码设备的普及,数字图像应用越来越广泛。而实际应用中,比如:人脸识别、OCR、图像分类、智能监控等,对图像的质量也有一定的要求,如何有效地评估一副图像质量的好坏已经变得越来越重要。目前的图像质量评估方法,无论是有参考的方式还是无参考的评估方式,都基本上是基于对整副图像进行处理,得到其对应的质量指标。事实上,计算机用这种方式进行质量评估时并没有真正体现出人眼视觉的处理过程。人眼视觉系统在评估周边场景时会实现自动对焦,这种对焦过程也就相当于寻找真正感兴趣的目标区域,也就是目标检测。图像中感兴趣区域(ROI)的质量对实际应用才更有意义。
因此,亟需一种图像质量评估方法和装置,使得图像质量评估过程更关注感兴趣的目标区域,忽略无关紧要的区域的图像质量。
发明内容
本发明针对上述问题,提出一种图像质量评估方法及装置。
本发明一方面提供了一种图像质量评估方法,所述方法包括:对待评估图像中的至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果;基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
在一种实施方式中,所述方法进一步包括:对所述待评估图像进行目标检测,以确定所述至少一个目标区域。
在一种实施方式中,对所述待评估图像进行目标检测的方法包括如下方法中的任一种:基于传统特征提取方式、基于深度特征提取方式。
在一种实施方式中,对所述待评估图像进行目标检测的步骤包括:对从所述待评估图像得到的二值化图像的轮廓进行分类,以确定文本轮廓;基于所确定的 文本轮廓,在指定方向上对文本进行合并,以确定所述至少一个目标区域。
在一种实施方式中,基于以下至少一个操作所确定的参数来对所述二值化图像的轮廓进行分类:确定所述二值化图像的轮廓内的非零像素个数;确定所述二值化图像的轮廓的高宽比以及宽高比;确定所述二值化图像的轮廓的指定方向范围内的指定方向邻域范围内存在的相似宽度轮廓的数目和相似高度轮廓的数目。
在一种实施方式中,基于所确定的文本轮廓,在指定方向上对文本进行合并的步骤包括:通过设置膨胀算子和腐蚀算子,在指定方向上对所确定的文本轮廓进行膨胀操作和腐蚀操作。
在一种实施方式中,对待评估图像中的至少一个目标区域分别进行质量评估的步骤包括:基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
在一种实施方式中,基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果的步骤包括:采用基于深度学习的训练方法确定图像质量评估模型,并基于所确定的所述至少一个目标区域和所述图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
在一种实施方式中,对所述至少一个目标区域分别进行质量评估的步骤包括:基于对像素灰度值的统计来对所述至少一个目标区域进行质量评估。
在一种实施方式中,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估的步骤包括:通过质量加权平均的方式,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
在一种实施方式中,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果的步骤包括:采用无参考质量评价指标,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果,其中,所述无参考质量评价指标包括边缘强度、噪声率或统一亮度分布中的至少一个。
本发明另一方面提供了一种图像质量评估装置,所述装置包括:目标区域质量评估单元,被配置为对待评估图像中的至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果;待评估图像质量评估单元,被配置为基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
在一种实施方式中,所述装置进一步包括:目标检测单元,被配置为对所述待评估图像进行目标检测,以确定所述至少一个目标区域。
在一种实施方式中,所述目标检测单元对所述待评估图像进行目标检测的方法包括如下方法中的任一种:基于传统特征提取方式、基于深度特征提取方式。
在一种实施方式中,所述目标检测单元包括:轮廓分类单元,被配置为从所 述待评估图像得到的二值化图像的轮廓进行分类,以确定文本轮廓;文本合并单元,被配置为基于所确定的文本轮廓,在指定方向上对文本进行合并,以确定所述至少一个目标区域。
在一种实施方式中,所述轮廓分类单元还被配置为:基于以下至少一个操作所确定的参数来对所述二值化图像的轮廓进行分类:确定所述二值化图像的轮廓内的非零像素个数;确定所述二值化图像的轮廓的高宽比以及宽高比;以及确定所述二值化图像的轮廓的指定方向范围内的指定方向邻域范围内存在的相似宽度轮廓的数目和相似高度轮廓的数目。
在一种实施方式中,所述文本合并单元还被配置为:通过设置膨胀算子和腐蚀算子,在指定方向上对所确定的文本轮廓进行膨胀操作和腐蚀操作。
在一种实施方式中,所述目标区域质量评估单元还被配置为:基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
在一种实施方式中,所述目标区域质量评估单元还被配置为:采用基于深度学习的训练方法确定图像质量评估模型,并基于所确定的所述至少一个目标区域和所述图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
在一种实施方式中,所述目标区域质量评估单元还被配置为:基于对像素灰度值的统计来对所述至少一个目标区域进行质量评估。
在一种实施方式中,所述待评估图像质量评估单元还被配置为:通过质量加权平均的方式,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
在一种实施方式中,所述目标区域质量评估单元还被配置为:采用无参考质量评价指标,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果,其中,所述无参考质量评价指标包括边缘强度、噪声率或统一亮度分布中的至少一个。
本发明的另一方面还提供了一种计算机设备,包括存储器、处理器以及存储在所述存储器上被所述处理器执行的计算机程序,所述处理器执行所述计算机程序时实现如上任一项所述的方法。
本发明的另一方面还提供了一种计算机存储介质,其上存储有处理器可执行程序,当所述处理器执行所述可执行程序时,实现如上任一项所述的方法。
本发明所提供的图像质量评估方法使得图像质量评估过程通过关注感兴趣的目标区域,忽略无关紧要的区域的图像质量,从而实现对待评估图像进行质量评估,评估速度快,评估精度较高,客观有效地评估了待评估图像的质量,使质量评估结果尽可能与人眼感知保持一致。
附图简要说明
图1是根据本发明实施例的图像质量评估方法的流程图。
图2是根据本发明实施例的图像质量评估另一方法的流程图。
图3是根据本发明实施例的文本图像质量评估方法的流程图。
图4是根据本发明实施例的一个文本图像的示例。
图5是图4的文本图像的二值化图像。
图6是对图4进行目标检测后得到的文本行区域的示意图。
图7是根据本发明实施例的图像质量评估模型的合成图像的一个示意图。
图8是图4的文本图像的评估结果。
图9是根据本发明实施例的另一个文本图像的示例。
图10是图9的文本图像的二值化图像。
图11是图9进行目标检测后得到的文本行区域的示意图。
图12是图9的文本图像的评估结果。
图13是根据本发明实施例的图像质量评估装置的示意图。
实施本发明的方式
为使本发明的目的、技术手段和优点更加清楚明白,以下结合附图对本发明作进一步详细说明。
图1是根据本发明实施例的图像质量评估方法的流程图。
S101:对待评估图像中的至少一个目标区域分别进行质量评估,以确定该至少一个目标区域中的每个目标区域的质量评估结果;
S102:基于所确定的该至少一个目标区域的每个目标区域的质量评估结果,对待评估图像进行质量评估。
为了评估图像质量,首先需要知道图像中的目标区域,本文中,确定目标区域的方式可以是预先设定,也可以是在每次评估前进行检测。
图2是根据本发明实施例的图像质量评估另一方法的流程图。
本发明提供了一种质量评估方法,该方法包括如图1所示的步骤:
S201:对待评估图像进行目标检测,以确定至少一个目标区域;
S202:对该至少一个目标区域分别进行质量评估,以确定该至少一个目标区域中的每个目标区域的质量评估结果;
S203:基于所确定的该至少一个目标区域的每个目标区域的质量评估结果,对待评估图像进行质量评估。
为了评估图像质量,首先需要检测图像中的目标区域。应理解的是,感兴趣的目标区域不同,采用目标检测方法不同。对待评估图像可以采用基于传统的特 征提取方式进行目标检测、基于深度特征提取方式进行目标检测(如CTPN方法、Haar分类器、Fater-rcnn)等,其中,待评估图像可以包括任意对象,譬如,动物、人脸、食品、汽车或文本行图像等。本文的待评估图像可以是彩色图像,也可以是灰度图像。进行目标检测之后返回的结果为多个感兴趣的目标区域,这些目标区域以图像的形式保存。
本文中,质量评估采用无参考质量评价指标,具体包括边缘强度、噪声率或统一亮度分布等。此外,在质量评估之前可以采用基于机器学习的方法训练质量评估模型,以用于对输入的待评估图像进行质量评估。模型训练需要预先收集各种目标区域图像,并进行质量标注,质量标注的大小,比如由数值0-100范围内的整数表示,标注数值越大表示对应图像质量越好,应理解,也可以采用其它合适的方式对质量标注的大小进行定义。在一种实施方式中,模型训练采用基于深度学习(卷积神经网络CNN)的训练方法,以确定较高准确度的图像质量模型。
此外,本文中,基于确定的至少一个目标区域中的每个目标区域的质量评估结果对待评估图像进行质量评估时,采用对至少一个目标区域中的每个目标质量区域的质量评估结果进行加权平均的方式,可以根据各个目标区域的重要性分别赋权值,目标区域的重要性取决于目标区域的大小、清晰度、对该目标区域的感兴趣程度等。当认为目标区域具有相同重要性时,加权平均的方式演变为各个目标区域的质量评估结果的算术平均。
以下结合附图对实施例作进行进一步详细的描述。
图3是根据本发明的文本图像质量评估方法的流程图,如图2所示,该方法包括如下步骤:
S301:输入待评估图像;
在该步骤中,待评估图像为文本图像(例如图4或图9所示),文本行是图像中感兴趣的目标,对文本行进行目标检测的目的是检测出文本行在图像中的位置。
S302:如果待评估图像是彩色图像,则执行步骤S303,否则跳过S303执行步骤S304;
S303:对彩色图像进行灰度化处理;
S304:对待处理灰度图像进行二值化处理;
在该步骤中,当输入的待评估图像为灰度图像(例如图4和图9所示)时,待评估图像即为待处理灰度图像,在执行完步骤S302后不经过步骤S303,直接到步骤S304,对待评估图像进行二值化处理;当输入的待评估图像为彩色图像时,待评估灰度图像为经过步骤S303灰度化处理的图像。本实施例中采用局部自适应二值化方法对待处理灰度图像进行二值化处理,即根据像素邻域的像素值分布来确定像素位置上的二值化阈值,对待处理灰度图像中的所有像素点,执行如下操作:
以像素点为中心,选取N×N邻域区域,在一种实施方式中,N在[2,5]之间;
计算该邻域范围内所有像素的均值;
将上步中计算得到的均值减去补偿常量Q,得到该像素的阈值,在一种实施方式中,Q在[2,7]之间;
将该像素值与上步求得的阈值作比较,若该像素值大于该阈值,则在图像中将该像素值设置为255,否则在图像中将该像素值设置为0。对图4和图9的图像进行二值化处理后,可得到对应的如图5和图10所示的二值化图像。
S305:对二值化图像的所有轮廓进行分类并去除非文本类轮廓,以确定文本轮廓,对二值化图像中的每一个轮廓进行如下操作:
确定轮廓范围内的非零像素值的个数nonz;
确定轮廓的高宽比hw以及宽高比wh;
确定轮廓的横向轮廓范围内横向邻域范围内(应理解,横向可以替换为纵向或任意合适的方向)存在的相似宽度轮廓数目SW和相似高度的轮廓数目SH;
若nonz小于第一阈值,和/或hw大于第二阈值并且SH小于第三阈值,和/或wh大于第四阈值且SW小于第五阈值,则认为上述轮廓皆为非文本轮廓,在二值化图像中将轮廓范围内所有位置的像素值置为0。在一种实施方式中,第一阈值在[2,5]之间,第二阈值在[8,12]之间,第三阈值在[2,5]之间,第四阈值在[8,12]之间,以及第五阈值在[2,5]之间。宽度为W并且高度为H的轮廓的相似宽度轮廓定义为该轮廓宽度范围在[0.7W,1.3W]之间,相似高度轮廓的轮廓高度在[0.7H,1.3H]之间。
S306:基于所确定的文本轮廓,对文本行进行合并,以确定文本行区域;
在该步骤中,具体地,将经步骤S305处理后的二值化图像中的轮廓在水平方向上(应理解,也可以在垂直方向上或任意合适的方向上)进行膨胀操作和腐蚀操作,得到文本行区域。其中,膨胀操作是对操作对象进行边界添加,而腐蚀则是删除对象边界的某些像素,其中,边界的定义由相应的操作算子给出,如膨胀算子大小为5×1,则以此像素为中心,将5×1的邻域范围内的像素均设置为目标像素,步骤S304、S305完成后,再利用大小为5×1的膨胀算子进行20次膨胀操作,以及利用大小为5×1的腐蚀算子进行15次腐蚀操作,以所有轮廓的外接矩形为掩码,可得到图像的文本行区域,应理解,在具体操作中,本领域技术人员可以对膨胀算子、膨胀操作次数和腐蚀次数做适当的调整。图6和图11为待评估图像完成步骤S306之后得到的文本行区域。
S307:对S306中检测到的文本行区域进行质量评估。
在质量评估步骤之前,采用基于深度学习的方法预先训练一个面向文本行区域的图像质量评估模型。其中,训练的文本行数据可以自己合成,也可以直接从文本图像中截取标注。自合成的文本行首先从常用的中文单字、英文单词以及中英文常用标点符号中随机选取候选字符组成字符串,然后将字符串与不同背景图 像相融合,再添加不同程度的模糊,再压缩成不同程度的质量并保存。根据合成图像的质量参数,标记对应图像的质量,大小为0-100之间,图7为合成图像的一个示意图。
分别将图6中的10个文本行区域和图11中的4个文本行区域输入到训练好的面向文本行区域的图像质量评估模型,得到相应的文本行区域质量评估结果。图8显示了图6的所有文本行区域line_1、line_2、line_3、line_4、line_5、line_6、line_7、line_8、line_9和line_10的质量评估结果,图12显示了图11的所有文本行区域line_1a、line_2a、line_3a和line_4a的质量评估结果。
S308:基于S307中得到的文本行区域的质量评估结果,通过质量加权平均的方式对待评估图像进行质量评估;
其中,加权值根据目标区域的重要性来确定,在本实施例中,假定目标区域的重要性相同,于是加权平均的方式简化为算术平均的方式,图4的文本图像的质量评估结果为图7所示的所有文本行区域line_1、line_2、line_3、line_4、line_5、line_6、line_7、line_8、line_9和line_10的质量评估结果的平均值,大小为16,图9的文本图像的质量评估结果为图12所示的所有文本行区域line_1a、line_2a、line_3a和line_4a的质量评估结果的平均值,大小为97。
通过上述描述可以看出,采用上述实施例的质量评估方法,关注感兴趣的目标区域,评估速度快,评估精度较高,有效地评估文本图像的质量,方便对质量不同的文本图像做后续处理。
在另一实施例中,质量评估的流程同样可以包括步骤S201、S202和S203。具体而言,在目标检测步骤中可以采用基于深度学习的算法。在目标区域质量评估的步骤中可以采用基于对像素灰度值的统计来对目标区域进行质量评估,例如,采用拉普拉斯方差算法。在对待评估图像进行质量评估的步骤中可以采用基于目标区域的质量评估结果,通过质量加权平均的方式对待评估图像进行质量评估。
图13根据本发明实施例的图像质量评估装置的示意图。
本发明还提供了如图13所示的一种质量评估装置1200,该装置包括目标检测单元1201、目标区域质量评估单元1202和待评估图像质量评估单元1203。具体地,目标检测单元1201被配置为对待评估图像进行检测,以确目标区域。目标区域质量评估单元1202被配置为对目标检测单元1201中确定的目标区域进行质量评估,以确定目标区域的质量评估结果。待评估图像质量评估单元1203被配置为基于目标区域质量评估单元1202中所确定的目标区域的质量评估结果,对待评估图像进行质量评估。譬如,待评估图像质量评估单元1203可以通过质量加权平均的方式对待评估图像进行质量评估。
此外,目标检测单元1201包括轮廓分类模块1202a和文本合并模块1202b。轮廓分类单元1202a被配置为基于以下至少一个操作所确定的参数来对所述二值 化图像的轮廓进行分类,以确定文本轮廓:确定二值化图像的轮廓内的非零像素个数nonz;确定二值化图像的轮廓的高宽比hw以及宽高比wh;以及确定二值化图像的轮廓的指定方向范围内的指定方向邻域范围内存在的相似宽度轮廓的数目SW和相似高度轮廓的数目SH。文本合并单元1202b还被配置为通过设置膨胀算子和腐蚀算子,在指定方向上对所确定的文本轮廓进行膨胀操作和腐蚀操作。文本合并单元1202b被配置为基于轮廓分类单元1202a中所确定的文本轮廓,在指定方向上对文本进行合并,以确定目标区域。
在一种实施方式中,目标区域质量评估单元1202还被配置为基于所确定的目标区域和图像质量评估模型,确定目标区域的质量评估结果。
在另一种实施方式中,目标区域质量评估单元1202还被配置为基于对像素灰度值的统计来对目标区域进行质量评估。
图1、2、3中的质量评估方法的流程还代表机器可读指令,该机器可读指令包括由处理器执行的程序。该程序可被实体化在被存储于有形计算机可读介质的软件中,该有形计算机可读介质如CD-ROM、软盘、硬盘、数字通用光盘(DVD)、蓝光光盘或其它形式的存储器。替代的,图1、2中的示例方法中的一些步骤或所有步骤可利用专用集成电路(ASIC)、可编程逻辑器件(PLD)、现场可编程逻辑器件(EPLD)、离散逻辑、硬件、固件等的任意组合被实现。另外,虽然图1、2所示的流程图描述了该质量评估方法,但可对该质量评估方法中的步骤进行修改、删除或合并。
如上所述,可利用编码指令(如计算机可读指令)来实现图1、2的示例过程,该编程指令存储于有形计算机可读介质上,如硬盘、闪存、只读存储器(ROM)、光盘(CD)、数字通用光盘(DVD)、高速缓存器、随机访问存储器(RAM)和/或任何其他存储介质,在该存储介质上信息可以存储任意时间(例如,长时间,永久地,短暂的情况,临时缓冲,和/或信息的缓存)。如在此所用的,该术语有形计算机可读介质被明确定义为包括任意类型的计算机可读存储的信号。附加地或替代地,可利用编码指令(如计算机可读指令)实现图1、2的示例过程,该编码指令存储于非暂时性计算机可读介质,如硬盘,闪存,只读存储器,光盘,数字通用光盘,高速缓存器,随机访问存储器和/或任何其他存储介质,在该存储介质信息可以存储任意时间(例如,长时间,永久地,短暂的情况,临时缓冲,和/或信息的缓存)。
以上仅为本发明的较佳实施例而已,并非用于限定本发明的保护范围。凡在本发明的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本发明的保护范围之内。

Claims (24)

  1. 一种图像质量评估方法,其特征在于,包括:
    对待评估图像中的至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果;
    基于所确定的所述至少一个目标区域中的质量评估结果,对所述待评估图像进行质量评估。
  2. 根据权利要求1所述的图像质量评估方法,其特征在于,进一步包括:
    对所述待评估图像进行目标检测,以确定所述至少一个目标区域。
  3. 根据权利要求2所述的图像质量评估方法,其特征在于,对所述待评估图像进行目标检测的方法包括如下方法中的任一种:
    基于传统特征提取方式、基于深度特征提取方式。
  4. 根据权利要求2或3所述的图像质量评估方法,其特征在于,对所述待评估图像进行目标检测的步骤包括:
    对从所述待评估图像得到的二值化图像的轮廓进行分类,以确定文本轮廓;
    基于所确定的文本轮廓,在指定方向上对文本进行合并,以确定所述至少一个目标区域。
  5. 根据权利要求4所述的图像质量评估方法,其特征在于,基于以下至少一个操作所确定的参数来对所述二值化图像的轮廓进行分类:
    确定所述二值化图像的轮廓内的非零像素个数;
    确定所述二值化图像的轮廓的高宽比以及宽高比;
    确定所述二值化图像的轮廓的指定方向范围内的指定方向邻域范围内存在的相似宽度轮廓的数目和相似高度轮廓的数目。
  6. 根据权利要求4或5所述的图像质量评估方法,其特征在于,基于所确定的文本轮廓,在指定方向上对文本进行合并的步骤包括:
    通过设置膨胀算子和腐蚀算子,在指定方向上对所确定的文本轮廓进行膨胀操作和腐蚀操作。
  7. 根据权利要求1至6中任一项所述的图像质量评估方法,其特征在于,对待评估图像中的至少一个目标区域分别进行质量评估的步骤包括:
    基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
  8. 根据权利要求7所述的图像质量评估方法,其特征在于,基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果的步骤包括:
    采用基于深度学习的训练方法确定图像质量评估模型,并基于所确定的所述至少一个目标区域和所述图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
  9. 根据权利要求1至8中任一项所述的图像质量评估方法,其特征在于,对所述至少一个目标区域分别进行质量评估的步骤包括:
    基于对像素灰度值的统计来对所述至少一个目标区域进行质量评估。
  10. 根据权利要求1至9中任一项所述的图像质量评估方法,其特征在于,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估的步骤包括:
    通过质量加权平均的方式,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
  11. 根据权利要求1至10中任一项所述的图像质量评估方法,其特征在于,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果的步骤包括:
    采用无参考质量评价指标,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果,
    其中,所述无参考质量评价指标包括边缘强度、噪声率或统一亮度分布中的至少一个。
  12. 一种图像质量评估装置,其特征在于,包括:
    目标区域质量评估单元,被配置为对待评估图像中的至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果;
    待评估图像质量评估单元,被配置为基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
  13. 根据权利要求12所述的图像质量评估装置,其特征在于,进一步包括:
    目标检测单元,被配置为对所述待评估图像进行目标检测,以确定所述至少一个目标区域。
  14. 根据权利要求13所述的图像质量评估装置,其特征在于,所述目标检测单元对所述待评估图像进行目标检测的方法包括如下方法中的任一种:
    基于传统特征提取方式、基于深度特征提取方式。
  15. 根据权利要求13或14所述的图像质量评估装置,其特征在于,所述目标检测单元包括:
    轮廓分类单元,被配置为对从所述待评估图像得到的二值化图像的轮廓进行分类,以确定文本轮廓;
    文本合并单元,被配置为基于所确定的文本轮廓,在指定方向上对文本进行合并,以确定所述至少一个目标区域。
  16. 根据权利要求15所述的图像质量评估装置,其特征在于,所述轮廓分类单元还被配置为:基于以下至少一个操作所确定的参数来对所述二值化图像的轮廓进行分类:
    确定所述二值化图像的轮廓内的非零像素个数;
    确定所述二值化图像的轮廓的高宽比以及宽高比;
    确定所述二值化图像的轮廓的指定方向范围内的指定方向邻域范围内存在的相似宽度轮廓的数目和相似高度轮廓的数目。
  17. 根据权利要求15或16所述的图像质量评估装置,其特征在于,所述文本合并单元还被配置为:通过设置膨胀算子和腐蚀算子,在指定方向上对所确定的文本轮廓进行膨胀操作和腐蚀操作。
  18. 根据权利要求12至17中任一项所述的图像质量评估装置,其特征在于,所述目标区域质量评估单元还被配置为:基于所确定的所述至少一个目标区域和图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
  19. 根据权利要求18中任一项所述的图像质量评估装置,其特征在于,所述目标区域质量评估单元还被配置为:
    采用基于深度学习的训练方法确定图像质量评估模型,并基于所确定的所述至少一个目标区域和所述图像质量评估模型,确定所述至少一个目标区域的质量评估结果。
  20. 根据权利要求12至19中任一项所述的图像质量评估装置,其特征在于,所述目标区域质量评估单元还被配置为:基于对像素灰度值的统计来对所述至少一个目标区域进行质量评估。
  21. 根据权利要求12至20中任一项所述的图像质量评估装置,其特征在于,所述待评估图像质量评估单元还被配置为:
    通过质量加权平均的方式,基于所确定的所述至少一个目标区域的质量评估结果,对所述待评估图像进行质量评估。
  22. 根据权利要求12至21中任一项所述的图像质量评估装置,其特征在于,所述目标区域质量评估单元还被配置为:采用无参考质量评价指标,对所述至少一个目标区域分别进行质量评估,以确定所述至少一个目标区域的质量评估结果,其中,所述无参考质量评价指标包括边缘强度、噪声率或统一亮度分布中的至少一个。
  23. 一种计算机设备,包括存储器、处理器以及存储在所述存储器上被所述处理器执行的计算机程序,其特征在于,所述处理器执行所述计算机程序时实现如权利要求1至11中任一项所述的方法。
  24. 一种计算机存储介质,其上存储有处理器可执行程序,当所述处理器执行所述可执行程序时,实现如上任一项所述的方法。
PCT/CN2018/106451 2017-09-20 2018-09-19 图像质量评估方法及装置 WO2019057067A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
SG11201907815V SG11201907815VA (en) 2017-09-20 2018-09-19 Method for assessing image quality and device thereof
JP2020504760A JP2020513133A (ja) 2017-09-20 2018-09-19 画像品質の評価方法及び装置

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710854415.9 2017-09-20
CN201710854415.9A CN107481238A (zh) 2017-09-20 2017-09-20 图像质量评估方法及装置

Publications (1)

Publication Number Publication Date
WO2019057067A1 true WO2019057067A1 (zh) 2019-03-28

Family

ID=60586643

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/106451 WO2019057067A1 (zh) 2017-09-20 2018-09-19 图像质量评估方法及装置

Country Status (4)

Country Link
JP (1) JP2020513133A (zh)
CN (1) CN107481238A (zh)
SG (1) SG11201907815VA (zh)
WO (1) WO2019057067A1 (zh)

Cited By (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110232381A (zh) * 2019-06-19 2019-09-13 梧州学院 车牌分割方法、装置、计算机设备及计算机可读存储介质
CN110236544A (zh) * 2019-05-29 2019-09-17 中国科学院重庆绿色智能技术研究院 基于相关系数的中风灌注成像病变区域检测系统及方法
CN111046886A (zh) * 2019-12-12 2020-04-21 吉林大学 号码牌自动识别方法、装置、设备及计算机可读存储介质
CN111275681A (zh) * 2020-01-19 2020-06-12 浙江大华技术股份有限公司 图片质量的确定方法及装置、存储介质、电子装置
CN111696083A (zh) * 2020-05-20 2020-09-22 平安科技(深圳)有限公司 一种图像处理方法、装置、电子设备及存储介质
CN112102309A (zh) * 2020-09-27 2020-12-18 中国建设银行股份有限公司 一种确定图像质量评估结果的方法、装置及设备
CN112287898A (zh) * 2020-11-26 2021-01-29 深源恒际科技有限公司 一种图像的文本检测质量评价方法及系统
CN112365451A (zh) * 2020-10-23 2021-02-12 微民保险代理有限公司 图像质量等级的确定方法、装置、设备及计算机可读介质
CN112767318A (zh) * 2020-12-31 2021-05-07 科大讯飞股份有限公司 一种图像处理效果的评价方法、装置、存储介质及设备
CN112801132A (zh) * 2020-12-28 2021-05-14 泰康保险集团股份有限公司 一种图像处理方法和装置
CN113450323A (zh) * 2021-06-22 2021-09-28 深圳盈天下视觉科技有限公司 质量检测方法、装置、电子设备及计算机可读存储介质
CN113627419A (zh) * 2020-05-08 2021-11-09 百度在线网络技术(北京)有限公司 兴趣区域评估方法、装置、设备和介质
JP2022519469A (ja) * 2020-01-02 2022-03-24 ▲広▼州大学 画像品質評価方法及び装置
CN115239961A (zh) * 2022-09-21 2022-10-25 江苏跃格智能装备有限公司 一种激光切割机工作状态监测方法
CN116246273A (zh) * 2023-03-07 2023-06-09 广州市易鸿智能装备有限公司 图像标注一致性评价方法、装置、电子设备及存储介质

Families Citing this family (27)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107481238A (zh) * 2017-09-20 2017-12-15 众安信息技术服务有限公司 图像质量评估方法及装置
CN108875731B (zh) * 2017-12-28 2022-12-09 北京旷视科技有限公司 目标识别方法、装置、系统及存储介质
CN108122231B (zh) * 2018-01-10 2021-09-24 山东华软金盾软件股份有限公司 监控视频下基于roi拉普拉斯算法的图像质量评价方法
CN111417981A (zh) * 2018-03-12 2020-07-14 华为技术有限公司 一种图像清晰度检测方法及装置
CN108446621A (zh) * 2018-03-14 2018-08-24 平安科技(深圳)有限公司 票据识别方法、服务器及计算机可读存储介质
CN108460766B (zh) * 2018-04-12 2022-02-25 四川和生视界医药技术开发有限公司 一种视网膜图像清晰度评估方法以及评估装置
CN108596084A (zh) * 2018-04-23 2018-09-28 宁波Gqy视讯股份有限公司 一种充电桩自动识别方法及装置
CN108805172A (zh) * 2018-05-08 2018-11-13 重庆瑞景信息科技有限公司 一种面向对象的图像效能盲评价方法
CN109104568A (zh) * 2018-07-24 2018-12-28 苏州佳世达光电有限公司 监控摄像头的智能清洁驱动方法及驱动系统
CN110874547B (zh) * 2018-08-30 2023-09-12 富士通株式会社 从视频中识别对象的方法和设备
CN111368837B (zh) * 2018-12-25 2023-12-05 中移(杭州)信息技术有限公司 一种图像质量评价方法、装置、电子设备及存储介质
CN109948625A (zh) * 2019-03-07 2019-06-28 上海汽车集团股份有限公司 文本图像清晰度评估方法及系统、计算机可读存储介质
CN110245577A (zh) * 2019-05-23 2019-09-17 复钧智能科技(苏州)有限公司 目标车辆识别方法、装置及车辆实时监控系统
CN110287826B (zh) * 2019-06-11 2021-09-17 北京工业大学 一种基于注意力机制的视频目标检测方法
CN110414519B (zh) * 2019-06-27 2023-11-14 众安信息技术服务有限公司 一种图片文字的识别方法及其识别装置、存储介质
CN112396574B (zh) * 2019-08-02 2024-02-02 浙江宇视科技有限公司 一种车牌图像质量处理方法、装置、存储介质及电子设备
CN110796624B (zh) * 2019-10-31 2022-07-05 北京金山云网络技术有限公司 一种图像生成方法、装置及电子设备
CN110827261B (zh) * 2019-11-05 2022-12-06 泰康保险集团股份有限公司 图像质量检测方法及装置、存储介质及电子设备
CN111192241B (zh) * 2019-12-23 2024-02-13 深圳市优必选科技股份有限公司 一种人脸图像的质量评估方法、装置及计算机存储介质
CN111595267B (zh) * 2020-05-18 2022-08-16 浙江大华技术股份有限公司 确定物体相位值的方法、装置、存储介质及电子装置
CN112396050B (zh) * 2020-12-02 2023-09-15 度小满科技(北京)有限公司 图像的处理方法、设备以及存储介质
CN112991313B (zh) * 2021-03-29 2021-09-14 清华大学 图像质量的评估方法及装置、电子设备和存储介质
CN113537192B (zh) * 2021-06-30 2024-03-26 北京百度网讯科技有限公司 图像检测方法、装置、电子设备及存储介质
CN113506260B (zh) * 2021-07-05 2023-08-29 贝壳找房(北京)科技有限公司 一种人脸图像质量评估方法、装置、电子设备及存储介质
CN113781428A (zh) * 2021-09-09 2021-12-10 广东电网有限责任公司 一种图像处理方法、装置、电子设备及存储介质
CN114067006B (zh) * 2022-01-17 2022-04-08 湖南工商大学 一种基于离散余弦变换的屏幕内容图像质量评价方法
CN114219803B (zh) * 2022-02-21 2022-07-15 浙江大学 一种三阶段图像质量评估的检测方法与系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049893A (zh) * 2011-10-14 2013-04-17 深圳信息职业技术学院 一种图像融合质量评价的方法及装置
CN106686377A (zh) * 2016-12-30 2017-05-17 佳都新太科技股份有限公司 一种基于深层神经网络的视频重点区域确定算法
CN107123122A (zh) * 2017-04-28 2017-09-01 深圳大学 无参考图像质量评价方法及装置
CN107481238A (zh) * 2017-09-20 2017-12-15 众安信息技术服务有限公司 图像质量评估方法及装置

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3400151B2 (ja) * 1994-12-08 2003-04-28 株式会社東芝 文字列領域抽出装置および方法
JP3710164B2 (ja) * 1995-05-02 2005-10-26 キヤノン株式会社 画像処理装置及び方法
JP2003208568A (ja) * 2002-01-10 2003-07-25 Ricoh Co Ltd 画像処理装置、画像処理方法、及び同方法に用いるプログラム
JP2007156741A (ja) * 2005-12-02 2007-06-21 Koito Ind Ltd 文字抽出方法、文字抽出装置およびプログラム
US9298979B2 (en) * 2008-01-18 2016-03-29 Mitek Systems, Inc. Systems and methods for mobile image capture and content processing of driver's licenses
CN101533474B (zh) * 2008-03-12 2014-06-04 三星电子株式会社 基于视频图像的字符和图像识别系统和方法
JP4821869B2 (ja) * 2009-03-18 2011-11-24 富士ゼロックス株式会社 文字認識装置、画像読取装置、およびプログラム
CN102054271B (zh) * 2009-11-02 2013-11-20 富士通株式会社 文本行检测方法和装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103049893A (zh) * 2011-10-14 2013-04-17 深圳信息职业技术学院 一种图像融合质量评价的方法及装置
CN106686377A (zh) * 2016-12-30 2017-05-17 佳都新太科技股份有限公司 一种基于深层神经网络的视频重点区域确定算法
CN107123122A (zh) * 2017-04-28 2017-09-01 深圳大学 无参考图像质量评价方法及装置
CN107481238A (zh) * 2017-09-20 2017-12-15 众安信息技术服务有限公司 图像质量评估方法及装置

Cited By (21)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110236544A (zh) * 2019-05-29 2019-09-17 中国科学院重庆绿色智能技术研究院 基于相关系数的中风灌注成像病变区域检测系统及方法
CN110236544B (zh) * 2019-05-29 2023-05-02 中国科学院重庆绿色智能技术研究院 基于相关系数的中风灌注成像病变区域检测系统及方法
CN110232381A (zh) * 2019-06-19 2019-09-13 梧州学院 车牌分割方法、装置、计算机设备及计算机可读存储介质
CN110232381B (zh) * 2019-06-19 2023-06-20 梧州学院 车牌分割方法、装置、计算机设备及计算机可读存储介质
CN111046886A (zh) * 2019-12-12 2020-04-21 吉林大学 号码牌自动识别方法、装置、设备及计算机可读存储介质
JP2022519469A (ja) * 2020-01-02 2022-03-24 ▲広▼州大学 画像品質評価方法及び装置
CN111275681A (zh) * 2020-01-19 2020-06-12 浙江大华技术股份有限公司 图片质量的确定方法及装置、存储介质、电子装置
CN111275681B (zh) * 2020-01-19 2023-09-01 浙江大华技术股份有限公司 图片质量的确定方法及装置、存储介质、电子装置
CN113627419A (zh) * 2020-05-08 2021-11-09 百度在线网络技术(北京)有限公司 兴趣区域评估方法、装置、设备和介质
CN111696083A (zh) * 2020-05-20 2020-09-22 平安科技(深圳)有限公司 一种图像处理方法、装置、电子设备及存储介质
CN111696083B (zh) * 2020-05-20 2024-05-14 平安科技(深圳)有限公司 一种图像处理方法、装置、电子设备及存储介质
CN112102309A (zh) * 2020-09-27 2020-12-18 中国建设银行股份有限公司 一种确定图像质量评估结果的方法、装置及设备
CN112365451A (zh) * 2020-10-23 2021-02-12 微民保险代理有限公司 图像质量等级的确定方法、装置、设备及计算机可读介质
CN112287898A (zh) * 2020-11-26 2021-01-29 深源恒际科技有限公司 一种图像的文本检测质量评价方法及系统
CN112801132A (zh) * 2020-12-28 2021-05-14 泰康保险集团股份有限公司 一种图像处理方法和装置
CN112767318A (zh) * 2020-12-31 2021-05-07 科大讯飞股份有限公司 一种图像处理效果的评价方法、装置、存储介质及设备
CN112767318B (zh) * 2020-12-31 2023-07-25 科大讯飞股份有限公司 一种图像处理效果的评价方法、装置、存储介质及设备
CN113450323A (zh) * 2021-06-22 2021-09-28 深圳盈天下视觉科技有限公司 质量检测方法、装置、电子设备及计算机可读存储介质
CN115239961A (zh) * 2022-09-21 2022-10-25 江苏跃格智能装备有限公司 一种激光切割机工作状态监测方法
CN116246273A (zh) * 2023-03-07 2023-06-09 广州市易鸿智能装备有限公司 图像标注一致性评价方法、装置、电子设备及存储介质
CN116246273B (zh) * 2023-03-07 2024-03-22 广州市易鸿智能装备有限公司 图像标注一致性评价方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
SG11201907815VA (en) 2019-11-28
CN107481238A (zh) 2017-12-15
JP2020513133A (ja) 2020-04-30

Similar Documents

Publication Publication Date Title
WO2019057067A1 (zh) 图像质量评估方法及装置
US10896349B2 (en) Text detection method and apparatus, and storage medium
WO2017121018A1 (zh) 二维码图像处理的方法和装置、终端、存储介质
US8902053B2 (en) Method and system for lane departure warning
KR102146560B1 (ko) 영상 보정 방법 및 장치
US9251614B1 (en) Background removal for document images
EP3467774A1 (en) Method for tracking and segmenting a target object in an image using markov chain, and device using the same
CN109255350B (zh) 一种基于视频监控的新能源车牌检测方法
CN105469027A (zh) 针对文档图像的水平和垂直线检测和移除
JP2018524732A (ja) 半自動画像セグメンテーション
US9275030B1 (en) Horizontal and vertical line detection and removal for document images
JP2006301847A (ja) 顔検出方法および装置並びにプログラム
US9171224B2 (en) Method of improving contrast for text extraction and recognition applications
JP6440278B2 (ja) 撮像装置及び画像処理方法
CN105303190B (zh) 一种基于对比度增强法的降质文档图像二值化方法
EP2821935A2 (en) Vehicle detection method and device
US8442348B2 (en) Image noise reduction for digital images using Gaussian blurring
CN117094975A (zh) 钢铁表面缺陷检测方法、装置及电子设备
US8300927B2 (en) Mouth removal method for red-eye detection and correction
CN104268845A (zh) 极值温差短波红外图像的自适应双局部增强方法
CN113537037A (zh) 路面病害识别方法、系统、电子设备及存储介质
CN110287752B (zh) 一种点阵码检测方法及装置
CN109948605B (zh) 一种针对小目标的图片增强方法及装置
JP2008165496A (ja) 画像正規化装置、対象物検出装置、対象物検出システム及びプログラム
US20190259168A1 (en) Image processing apparatus, image processing method, and storage medium

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18857750

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020504760

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 16.09.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18857750

Country of ref document: EP

Kind code of ref document: A1