WO2018192245A1 - 一种基于美学评价的照片自动评分方法 - Google Patents

一种基于美学评价的照片自动评分方法 Download PDF

Info

Publication number
WO2018192245A1
WO2018192245A1 PCT/CN2017/116059 CN2017116059W WO2018192245A1 WO 2018192245 A1 WO2018192245 A1 WO 2018192245A1 CN 2017116059 W CN2017116059 W CN 2017116059W WO 2018192245 A1 WO2018192245 A1 WO 2018192245A1
Authority
WO
WIPO (PCT)
Prior art keywords
photo
aesthetic
image
model
close
Prior art date
Application number
PCT/CN2017/116059
Other languages
English (en)
French (fr)
Inventor
刘弋锋
吕相文
陈洛奇
谢海永
Original Assignee
中国电子科技集团公司电子科学研究院
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
Priority claimed from CN201710257193.2A external-priority patent/CN107153838A/zh
Priority claimed from CN201710257194.7A external-priority patent/CN107018330A/zh
Application filed by 中国电子科技集团公司电子科学研究院 filed Critical 中国电子科技集团公司电子科学研究院
Publication of WO2018192245A1 publication Critical patent/WO2018192245A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition

Definitions

  • the present invention relates to the field of image processing technologies, and in particular, to a photo automatic scoring method based on aesthetic evaluation.
  • Apple's iOS10 system has been able to automatically classify based on the identified photo content.
  • the invention with the application number CN201510827259.8 performs image recognition on the photos in the album, classifies the photos as portrait categories or landscape categories, acquires the shooting information of the photos in the album, and classifies the photos of the portrait categories according to the shooting information.
  • the self-portrait portrait category or his portrait category is used to classify photos of the landscape category into day or night categories.
  • the invention classifies the photo content by real-time intelligent recognition of the photo, so that the photo album is more tidy, but in the face of the massive quality of the jagged photos, the classification of the photo quality from the aesthetic point of view cannot be realized.
  • the invention with the application number CN201210359524.0 first extracts the subject area of the sample photo based on the power spectrum slope method, and then extracts the feature of the sample photo, and finally uses the support vector machine classifier to train the photo aesthetic quality to obtain the interface model;
  • the automatic classification process uses the interface model for identification.
  • the invention considers the idea of classification from the aesthetic point of view, but its artificially extracted features are not comprehensive, have certain subjectivity and one-sidedness, and adopt a general discriminant strategy without distinguishing the photo scenes, due to the aesthetics of different photo scenes.
  • the invention of CN201080042531.7 determines one or more vanishing points associated with an input digital image by automatically analyzing the digital image; calculating a component model based on at least the position of the vanishing point; generating an input digital map in response to the component model
  • the aesthetic quality parameter is an estimate of the aesthetic quality of the input digital image.
  • the invention uses manually set rules and does not take into account the differences in evaluation criteria for different scenarios.
  • an automatic photo-rating method based on aesthetic evaluation capable of automatically adapting to different image scenes and automatically adapting to landscape images and person images is needed.
  • the first information of the photo is input into the aesthetic evaluation model for scoring
  • the second information of the photo is input into the aesthetic evaluation model for scoring.
  • the first information includes an image scene classification result and the image itself.
  • the aesthetic evaluation model comprises a single model and a multi-model, and the aesthetic evaluation model is obtained by machine learning training, wherein the training method of the aesthetic evaluation model comprises:
  • a plurality of models corresponding to the scenes are respectively trained on the preset aesthetic evaluation models according to different scene labels to which the images belong.
  • the photo when the photo is a non-close-up photo, if the photo can read the EXIF information, the first information includes an image scene classification result, an image photo EXIF, and the image itself.
  • the aesthetic evaluation model comprises a single model and a multi-model, and the aesthetic evaluation model is obtained by machine learning training, wherein the training method of the aesthetic evaluation model comprises:
  • the predetermined aesthetic evaluation model is trained by using the known aesthetic score, the image scene category label, and the EXIF information to obtain the single model;
  • a plurality of models corresponding to the scenes are respectively trained on the preset aesthetic evaluation models according to different scene labels to which the images belong.
  • the location coordinates of the region of interest of the photo is extracted; if the photo cannot read the EXIF information, the second information includes an image scene classification result, a location coordinates of the region of interest, The image itself.
  • the aesthetic evaluation model comprises a single model and a multi-model, and the aesthetic evaluation model is obtained by machine learning training, wherein the training method of the aesthetic evaluation model comprises:
  • a plurality of models corresponding to the plurality of scenes are respectively trained on the preset aesthetic evaluation model according to different scene labels to which the images belong.
  • the location coordinates of the region of interest of the photo is extracted; if the photo can read the EXIF information, the second information includes an image scene classification result, an image photo EXIF, and an interest Area position coordinates, the image itself.
  • the aesthetic evaluation model comprises a single model and a multi-model, and the aesthetic evaluation model is obtained by machine learning training, wherein the training method of the aesthetic evaluation model comprises:
  • the predetermined aesthetic evaluation model is trained by using the known aesthetic score, the image scene category label, the region of interest location coordinates, and the EXIF information to obtain the single model;
  • the preset aesthetic evaluation models are separately trained according to the scene labels to which the images belong. Multiple models corresponding to multiple scenes.
  • whether the photo is a close-up photo is judged by the following method:
  • the photo is a close-up photo, otherwise it is a non-close-up photo.
  • the invention adopts an image scene classification algorithm to adaptively evaluate the different scenes, and avoids the classification performance limitation of the general aesthetic discriminant strategy, and improves the accuracy of the final album photo classification.
  • the invention utilizes the photo EXIF information, can more accurately obtain various parameters of the photo shooting, and ensures that the aesthetic score is more accurate.
  • the invention utilizes the position information of the close-up main body of the image, and takes into account the composition factor of the main body, so the aesthetic score of the photo image mainly composed of the above object is more accurate.
  • the aesthetic evaluation model in the present invention is trained by the machine learning method, and the subjectivity and one-sidedness of the manual rule evaluation are avoided.
  • FIG. 1 is a schematic diagram showing a computing device for an automatic photo rating method based on aesthetic evaluation of the present invention.
  • FIG. 2 is a flow chart showing a method for automatically scoring a photo based on aesthetic evaluation according to the present invention
  • FIG. 3 is a flow chart showing a single model training for evaluating non-close-up photographs in which EXIF information cannot be read;
  • FIG. 4 is a flow chart showing a multi-model training for evaluating non-close-up photographs in which EXIF information cannot be read;
  • Figure 5 is a flow chart showing a single model training for evaluating non-close-up photographs capable of reading EXIF information
  • Figure 6 shows the evaluation of the present invention for evaluating non-close-up photographs capable of reading EXIF information. Multi-model training flow chart;
  • Figure 7 is a flow chart showing a single model training for evaluating close-up photographs of EXIF information that cannot be read by the present invention
  • Figure 8 is a flow chart showing the multi-model training of the present invention for evaluating close-up photographs in which EXIF information cannot be read;
  • Figure 9 is a flow chart showing a single model training for evaluating close-up photographs of EXIF information of the present invention.
  • Figure 10 is a flow chart showing the multi-model training of the present invention for evaluating close-up photographs capable of reading EXIF information
  • Figure 11 shows a single model aesthetic score flow chart for a non-close-up photo that cannot read EXIF information
  • Figure 12 shows a multi-model aesthetic score flow chart for a non-close-up photo that cannot read EXIF information
  • Figure 13 shows a single model aesthetic score flow chart for non-close-up photographs that can read EXIF information
  • Figure 14 shows a multi-model aesthetic score flow chart for non-close-up photographs that can read EXIF information
  • Figure 15 shows a single model aesthetic score flow chart for a close-up photo of the EXIF information
  • Figure 16 shows a multi-model aesthetic score flow chart for a close-up photo of EXIF information
  • Figure 17 shows a single model aesthetic score flow chart for reading close-up photos of EXIF information
  • Figure 18 shows a multi-model aesthetic score flow chart for a close-up photo of EXIF information.
  • FIG. 1 is a schematic diagram of a computing device for an esthetic evaluation-based photo automatic scoring method according to the present invention.
  • an exemplary mobile phone is used as a carrier, and a large number of photos 101 are stored in a computing device, and a processor chip of the computing device is executed.
  • the evaluation of the beauty evaluation model, as well as the acquisition of a large number of photos stored in the memory, automatically score the photos.
  • the computing device can be a camera that takes images directly or other smart devices known to those skilled in the art.
  • FIG. 2 is a flow chart of a method for automatically scoring a photo based on aesthetic evaluation according to the present invention.
  • a smart phone is taken as an example, and an automatic photo-rating method based on aesthetic evaluation provided by the present invention is implemented in a computing device. And execute the following instructions:
  • the first information of the photo is input into the aesthetic evaluation model for scoring.
  • the second information of the photo is input into the aesthetic evaluation model for scoring.
  • the aesthetic evaluation model of the present invention includes a single model and a multi-model, and the aesthetic evaluation model classifies whether the photos to be scored are close-up photos, and the training process passes the machine learning method.
  • the method is trained, and the machine learning methods used include convolutional neural networks, restricted Boltzmann machines, deep confidence networks, and the like.
  • the aesthetic evaluation model is trained to obtain a single model and a multi-model, and the aesthetic evaluation model is obtained through machine learning training.
  • the present invention is a single model training flowchart for evaluating non-close-up photographs in which EXIF information cannot be read, and a single model training for evaluating non-close-up photographs in which EXIF information cannot be read is trained as follows. :
  • the present invention is used for a multi-model training flowchart for evaluating non-close-up photographs in which EXIF information cannot be read, and multi-model training for evaluating non-close-up photographs in which EXIF information cannot be read is trained as follows:
  • the plurality of models corresponding to the scene are respectively trained on the preset aesthetic evaluation model according to different scene labels to which the images belong.
  • Exemplary models of landscape images and plant image images in the embodiment are used to obtain a model for aesthetic evaluation of landscape images and a model for aesthetic evaluation of plant images for multi-model training, and those skilled in the art should understand that it is not limited thereto.
  • the aesthetic evaluation model is trained to obtain a single model and a multi-model for the photo that can read the EXIF information.
  • the present invention is a single model training flowchart for evaluating non-close-up photographs capable of reading EXIF information, and a single model training for evaluating non-close-up photographs capable of reading EXIF information is trained as follows. :
  • the present invention is a multi-model training flowchart for evaluating non-close-up photographs capable of reading EXIF information
  • multi-model training for evaluating non-close-up photographs capable of reading EXIF information is trained as follows. :
  • the plurality of models corresponding to the scene are respectively trained on the preset aesthetic evaluation model according to different scene labels to which the image belongs.
  • Exemplary models of landscape images and plant image images in the embodiment are used to obtain a model for aesthetic evaluation of landscape images and a model for aesthetic evaluation of plant images for multi-model training, and those skilled in the art should understand that it is not limited thereto.
  • the EXIF information in the embodiment of the present invention includes information such as aperture value, shutter value, focal length, exposure time, ISO, etc., and those skilled in the art should understand that it is not limited thereto.
  • the aesthetic evaluation model is trained to obtain a single model and a multi-model, and the aesthetic evaluation model is obtained through machine learning training.
  • the present invention is a single model training flowchart for evaluating close-up photographs in which EXIF information cannot be read, and a single model training for evaluating close-up photographs in which EXIF information cannot be read is trained as follows:
  • S110 Training a preset aesthetic evaluation model by using a plurality of images of known aesthetic scores, image scene category labels, and location coordinates of the region of interest to obtain a single model.
  • the present invention is a multi-model training flowchart for evaluating close-up photographs in which EXIF information cannot be read, and multi-model training for evaluating close-up photographs in which EXIF information cannot be read is trained as follows:
  • S111 Acquire a plurality of images of known aesthetic scores, image scene category labels, region of interest location coordinates, and EXIF information.
  • a plurality of models corresponding to multiple scenes are respectively trained on the preset aesthetic evaluation model.
  • Exemplary models of landscape images and plant image images in the embodiment are used to obtain a model for aesthetic evaluation of landscape images and a model for aesthetic evaluation of plant images for multi-model training, and those skilled in the art should understand that it is not limited thereto.
  • the aesthetic evaluation model is trained to obtain a single model and multiple models.
  • the present invention is a single model training flowchart for evaluating close-up photographs of EXIF information, and a single model training for evaluating close-up photographs of EXIF information can be trained as follows:
  • the present invention is a multi-model training flowchart for evaluating close-up photographs of EXIF information
  • a multi-model training for evaluating close-up photographs of EXIF information can be trained as follows:
  • the plurality of models corresponding to the plurality of scenarios are respectively trained on the preset aesthetic evaluation model according to different scene labels to which the image belongs.
  • Exemplary models of landscape images and plant image images in the embodiment are used to obtain a model for aesthetic evaluation of landscape images and a model for aesthetic evaluation of plant images for multi-model training, and those skilled in the art should understand that it is not limited thereto.
  • the image scene classification algorithm is used to perform image recognition and scene classification on the photos acquired by the smart device, and determine whether the photo is a close-up photo.
  • the source of the image of the present invention can It is a photo stored in the computing device, and the smart device camera live view image can also be obtained as the image source.
  • the image scene classification algorithm is used to identify the photos, and the photos are divided into landscape, night scene, architecture, dynamic, static, backlight, portrait, animal, and plant according to the recognition result.
  • the scene classification algorithm includes the following steps to implement scene classification of the photo:
  • the image can be classified into multiple categories of landscape, night scene, architecture, dynamic, static, backlight, portrait, animal, and plant, that is, the scene classification result of the image is obtained.
  • the photo When the area/area ratio of the most significant area is greater than the threshold, the photo is a close-up photo, otherwise it is a non-close-up photo.
  • the single-model aesthetic score flow chart of the non-close-up photo of the EXIF information cannot be read.
  • the first information of the photo includes the image scene classification result and the image itself. .
  • the first information of the photo is input to the single model obtained in step S102 for scoring, and an aesthetic score of the image is obtained. It should be understood that the result of the scene classification in the input single model is a category label of a different class of the image scene.
  • the multi-model aesthetic score flow chart of the non-close-up photo of the EXIF information cannot be read.
  • the first information of the photo includes the image scene classification result and the image itself.
  • the images of the different types of scenes are input into step S104 to obtain an aesthetic evaluation model of the multi-model corresponding scene, and the aesthetic score of the image is obtained.
  • exemplary landscape image photographs and plant image photographs are used for scoring, and the landscape image photographs aesthetic scores and plants are obtained respectively.
  • the class image photo aesthetic score which one skilled in the art should understand, is not limited thereto.
  • the single-model aesthetic score flow chart of the non-close-up photo of the EXIF information can be read.
  • the first information includes the image scene classification result and the image photo EXIF. , the image itself.
  • the first information of the photo is input to the single model obtained in step S106 to perform scoring, and an aesthetic score of the image is obtained. It should be understood that the result of the scene classification in the input single model is a category label of a different class of the image scene.
  • a multi-model aesthetic scoring flowchart capable of reading non-close-up photos of EXIF information
  • the first information of the photo includes image scene classification result, image photo EXIF and the image itself.
  • the scene classification result of the image the image photo EXIF of the different category scenes and the image itself are input into step S108 to obtain an aesthetic evaluation model of the multi-model corresponding scene, and the aesthetic score of the image is obtained.
  • the landscape image photograph and the plant image photograph are exemplified to obtain a landscape image photograph aesthetic score and a plant image photograph aesthetic score, respectively, and those skilled in the art understand that it is not limited thereto.
  • the photo is taken as a close-up photo, and the coordinates of the region of interest of the photo are extracted.
  • the position coordinates of the region of interest of the extracted photograph are used to locate the salient region by the saliency detection algorithm, and the positioning is performed.
  • the significant area position is used as the coordinates of the area of interest.
  • the region of interest location coordinates may employ a manually specified region location as the region of interest location coordinates.
  • the single-model aesthetic score flow chart of the close-up photo of the EXIF information cannot be read.
  • the second information includes the image scene classification result, the coordinates of the region of interest, and The image itself.
  • the second information of the photo is input to the single model obtained in step S110 for scoring, and an aesthetic score of the image is obtained. It should be understood that the result of the scene classification in the input single model is a category label of a different class of the image scene.
  • the multi-model aesthetic scoring flowchart of the close-up photo of the EXIF information cannot be read.
  • the second information of the photo includes the image scene classification result and the location of the region of interest. The coordinates and the image itself.
  • the region coordinates of the region of interest of different types of scenes and the image itself are input into step S112 to obtain an aesthetic evaluation model of the multi-model corresponding scene, and the aesthetic score of the image is obtained.
  • Exemplary in the examples are scored by landscape image photos and plant image images.
  • the landscape image photo aesthetic score and the plant image photo aesthetic score are respectively obtained, and those skilled in the art should understand that it is not limited thereto.
  • a single model aesthetic score flow chart capable of reading a close-up photo of EXIF information.
  • the second information includes image scene classification result, image photo EXIF information, and sense. Interest area location coordinates and the image itself.
  • the second information of the photo is input to the single model obtained in step S110 for scoring, and an aesthetic score of the image is obtained. It should be understood that the result of the scene classification in the input single model is a category label of a different class of the image scene.
  • a multi-model aesthetic score flow chart capable of reading a close-up photo of EXIF information.
  • the second information of the photo includes image scene classification result and image photo EXIF information. , the location coordinates of the region of interest and the image itself.
  • the scene classification result of the image the region coordinates of interest of the different types of scenes and the image itself are input into step S114 to obtain an aesthetic evaluation model of the multi-model corresponding scene, and the aesthetic score of the image is obtained.
  • the landscape image photograph and the plant image photograph are exemplified to obtain a landscape image photograph aesthetic score and a plant image photograph aesthetic score, respectively, and those skilled in the art understand that it is not limited thereto.
  • the photo automatic scoring method based on the aesthetic evaluation of the present invention can be applied to automatic photo grading, landscape and human shooting guidance, and aesthetic style recommendation.
  • the photo automatic scoring method based on the aesthetic evaluation of the present invention can quantitatively score the beauty of the photo album photos in the smart device, and divide the scoring range by the scoring method obtained by applying the scoring method of the present invention, and divide the scoring results into several grades to realize Album photos are automatically graded from an aesthetic perspective and automatically categorized by level.
  • the image-based automatic scoring method based on the aesthetic evaluation of the present invention can quantitatively score the images of the live view in the smart device, and helps guide the user to take pictures of landscapes and people.
  • the image photo automatic scoring method based on the aesthetic evaluation of the present invention can recommend an image photograph conforming to the aesthetic feeling to the user according to the scoring result of the image photograph in the user equipment.

Abstract

本发明提供了一种基于美学评价的照片自动评分方法，所述方法包括：训练美学评价模型；获取智能设备中的照片，采用图像场景分类算法对所述照片进行图像识别并进行场景分类，并判断所述照片是否为特写照片；当所述照片为非特写照片时，将所述照片的第一信息输入到所述美学评价模型中进行评分；当所述照片为特写照片时，将所述照片的第二信息输入到所述美学评价模型中进行评分。本发明通过训练美学评价模型，通过场景分类算法将获取的照片进行场景分类后引入到美学评价模型对图像进行自动评分，提高评分的准确性，避免了人工规则评价的主观性和片面性。

Description

一种基于美学评价的照片自动评分方法
专利的交叉引用
本申请要求2017年04月19日提交的，申请号CN201710257193.2、申请号CN201710257194.7的中国发明专利申请的优选权。
技术领域
本发明涉及图像处理技术领域，特别涉及一种基于美学评价的照片自动评分方法。
背景技术
现今用户在个人设备中存储的照片越来越多，因此需要一种能对照片进行组织管理的方法。目前已有一些可以对图像照片进行组织管理的应用和发明专利。苹果公司iOS10系统已经可以根据识别出的照片内容进行自动分类。申请号为CN201510827259.8的发明对相册中的相片进行图像识别，将相片归类为人像类别或风景类别；获取相册中相片的拍摄信息，根据所述拍摄信息将人像类别的相片再归类为自拍人像类别或他拍人像类别，将风景类别的相片再归类为日景类别或夜景类别。该发明通过对相片的实时智能识别来进行相片内容的分类，使得相册更加整洁，但是面对海量质量参差不齐的照片，无法实现从美学角度进行照片质量的分类。申请号为CN201210359524.0的发明先采用基于功率谱斜度的方法提取样本照片的主题区域，然后提取样本照片的特征，最终利用支持矢量机分类器进行照片美学质量的训练，得到分界面模型；自动分类过程使用分界面模型进行识别。该发明考虑了从美学角度分类的思路，但是其通过人工提取的特征并不全面，具有一定的主观性和片面性，且采取通用判别策略而没有针对照片场景进行区别处理，由于不同照片场景的美学评价标准应有所区别，因此分类性能有限。CN201080042531.7的发明通过自动分析数字图像确定与输入数字图像相关的一个或多个灭点；至少根据灭点的位置计算组分模型；回应组分模型产生输入数字图 像的美学质量参数，其中，美学质量参数是输入数字图像的美学质量的估计。该发明采用人工设定的规则，同样也没有考虑到不同场景的评价标准区别。
因此，为了解决上述问题，需要能够自动适应不同的图像场景，且能自动适应风景图像和人物图像的一种基于美学评价的照片自动评分方法。
发明内容
本发明的目的在于提供一种基于美学评价的照片自动评分方法，所述方法包括在计算设备中执行的如下指令：
训练美学评价模型；
获取智能设备中的照片，采用图像场景分类算法对所述照片进行图像识别并进行场景分类，并判断所述照片是否为特写照片；
当所述照片为非特写照片时，将所述照片的第一信息输入到所述美学评价模型中进行评分；
当所述照片为特写照片时，将所述照片的第二信息输入到所述美学评价模型中进行评分。
优选地，所述照片为非特写照片时，若所述照片不能读取EXIF信息，则所述第一信息包括图像场景分类结果和图像本身。
优选地，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像；
利用所述已知美学评分和图像场景类别标签的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
利用已知美学评分的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。
优选地，所述照片为非特写照片时，若所述照片可以读取EXIF信息，则所述第一信息包括图像场景分类结果、图像照片EXIF、图像本身。
优选地，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像；
利用所述已知美学评分、图像场景类别标签和EXIF信息的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
利用已知美学评分和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。
优选地，所述照片为特写照片时，提取所述照片的感兴趣区域位置坐标；若所述照片不能读取EXIF信息，则所述第二信息包括图像场景分类结果、感兴趣区域位置坐标、图像本身。
优选地，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像；
利用所述已知美学评分、图像场景类别标签和感兴趣区域位置坐标的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
利用已知美学评分和感兴趣区域位置坐标的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与多个场景对应的多模型。
优选地，所述照片为特写照片时，提取所述照片的感兴趣区域位置坐标；若所述照片可以读取EXIF信息，则所述第二信息包括图像场景分类结果、图像照片EXIF、感兴趣区域位置坐标、图像本身。
优选地，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像；
利用所述已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
利用已知美学评分、感兴趣区域位置坐标和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练 出与多个场景对应的多模型。
优选地，所述照片是否为特写照片通过如下方法判断：
通过显著性检测算法检测照片中的显著性区域；
当最大显著性区域面积/面积比例大于阈值时，则所述照片为特写照片，否则为非特写照片。
本发明采用图像场景分类算法对不同场景进行自适应地美学评价，而且避免采取通用美学判别策略的分类性能限制，提高了最终相册照片分级的准确性。
本发明利用了照片EXIF信息，可以更准确的获取照片拍摄时的各项参数，保证美学评分更加准确。
本发明利用了图像特写主体物体位置信息，考虑到了主体的构图因素，因此对以上述物体为主体的照片图像的美学评分更加准确。
本发明中的美学评价模型通过机器学习方法训练得到，避免了人工规则评价的主观性和片面性。
应当理解，前述大体的描述和后续详尽的描述均为示例性说明和解释，并不应当用作对本发明所要求保护内容的限制。
附图说明
参考随附的附图，本发明更多的目的、功能和优点将通过本发明实施方式的如下描述得以阐明，其中：
图1示意性示出了用于本发明基于美学评价的照片自动评分方法的计算设备示意图。
图2示出了本发明基于美学评价的照片自动评分方法的流程框图；
图3示出了本发明用于对不能读取EXIF信息的非特写照片进行评价的单模型训练流程图；
图4示出了本发明用于对不能读取EXIF信息的非特写照片进行评价的多模型训练流程图；
图5示出了本发明用于对能读取EXIF信息的非特写照片进行评价的单模型训练流程图；
图6示出了本发明用于对能读取EXIF信息的非特写照片进行评价的 多模型训练流程图；
图7示出了本发明用于对不能读取EXIF信息的特写照片进行评价的单模型训练流程图；
图8示出了本发明用于对不能读取EXIF信息的特写照片进行评价的多模型训练流程图；
图9示出了本发明用于对能读取EXIF信息的特写照片进行评价的单模型训练流程图；
图10示出了本发明用于对能读取EXIF信息的特写照片进行评价的多模型训练流程图；
图11示出了不能读取EXIF信息的非特写照片的单模型美学评分流程图；
图12示出了不能读取EXIF信息的非特写照片的多模型美学评分流程图；
图13示出了能读取EXIF信息的非特写照片的单模型美学评分流程图；
图14示出了能读取EXIF信息的非特写照片的多模型美学评分流程图；
图15示出了不能读取EXIF信息的特写照片的单模型美学评分流程图；
图16示出了不能读取EXIF信息的特写照片的多模型美学评分流程图；
图17示出了能读取EXIF信息的特写照片的单模型美学评分流程图；
图18示出了能读取EXIF信息的特写照片的多模型美学评分流程图。
具体实施方式
通过参考示范性实施例，本发明的目的和功能以及用于实现这些目的和功能的方法将得以阐明。然而，本发明并不受限于以下所公开的示范性实施例；可以通过不同形式来对其加以实现。说明书的实质仅仅是帮助相关领域技术人员综合理解本发明的具体细节。
在下文中，将参考附图描述本发明的实施例，相关技术术语应当是本领域技术人员所熟知的。在附图中，相同的附图标记代表相同或类似 的部件，或者相同或类似的步骤，除非另有说明。下面通过具体的实施例对本发明的内容进行说明。
在下文中解决具体的实施例对本发明提供的一种基于美学评价的照片自动评分方法进行详细说明，现有技术中对照片图像的处理缺乏根据不同场景对照片图像进行评分，使得在面对海量质量参差不齐的照片时，无法实现从美学角度进行照片质量的分类。
本发明提供的一种基于美学评价的照片自动评分方法执行在计算设备，计算设备包括但不限于个人个人电脑、智能手机、平板电脑、智能眼镜，以及其他具有摄像头、数据存储和数据处理功能的设备。如图1所示用于本发明基于美学评价的照片自动评分方法的计算设备示意图，实施例中示例性的以手机作为载体，在计算设备中存储器存储大量照片101，计算设备的处理器芯片执行美评价模型的训练，以及获取存储在存储器中的大量照片，对照片进行自动评分。在一些实施例中，计算设备可以是直接进行拍摄图像的相机或其他本领域技术人员所公知的智能设备。
如图2所示本发明基于美学评价的照片自动评分方法的流程框图，本发明的一个实施例中以智能手机为例，本发明提供的一种基于美学评价的照片自动评分方法执行在计算设备，并执行如下指令：
S1、训练美学评价模型。
S2、获取智能设备中的照片。
S3、采用图像场景分类算法对所述照片进行图像识别并进行场景分类。
S4、判断所述照片是否为特写照片：
当所述照片为非特写照片时，将所述照片的第一信息输入到所述美学评价模型中进行评分。
当所述照片为特写照片时，将所述照片的第二信息输入到所述美学评价模型中进行评分。
本发明的美学评价模型包括单模型和多模型，美学评价模型针对需要评分的照片是否为特写照片进行分类训练，训练过程通过机器学习方 法进行训练得到，使用的机器学习方法包括卷积神经网络、受限玻尔兹曼机、深度置信网络等。
具体地，当需要评分的照片为非特写照片时，针对不能读取EXIF信息的照片，美学评价模型训练后得到单模型和多模型，美学评价模型通过机器学习训练得到。
如图3所示本发明用于对不能读取EXIF信息的非特写照片进行评价的单模型训练流程图，用于对不能读取EXIF信息的非特写照片进行评价的单模型训练按照如下方法训练：
S101、获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像。
S102、利用已知美学评分和图像场景类别标签的若干个图像对预设的美学评价模型进行训练，得到单模型。
如图4所示本发明用于对不能读取EXIF信息的非特写照片进行评价的多模型训练流程图，对不能读取EXIF信息的非特写照片进行评价的多模型训练按照如下方法训练：
S103、获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像。
S104、利用已知美学评分的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。实施例中示例性的以风景类图像照片和植物类图像照片为了进行多模型训练分别得到风景类图像美学评价的模型和植物类图像美学评价的模型，本领域技术人员应当理解并不限于此。
当需要评分的照片为非特写照片时，针对能读取EXIF信息的照片，美学评价模型训练后得到单模型和多模型。如图5所示本发明用于对能读取EXIF信息的非特写照片进行评价的单模型训练流程图，用于对能读取EXIF信息的非特写照片进行评价的单模型训练按照如下方法训练：
S105、获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像。
S106、利用所述已知美学评分、图像场景类别标签和EXIF信息的若 干个图像对预设的美学评价模型进行训练，得到单模型。
如图6所示本发明用于对能读取EXIF信息的非特写照片进行评价的多模型训练流程图，用于对能读取EXIF信息的非特写照片进行评价的多模型训练按照如下方法训练：
S107、获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像。
S108、利用已知美学评分和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。实施例中示例性的以风景类图像照片和植物类图像照片为了进行多模型训练分别得到风景类图像美学评价的模型和植物类图像美学评价的模型，本领域技术人员应当理解并不限于此。
本发明的实施例中EXIF信息包括光圈值、快门值、焦距、曝光时间、ISO等信息，本领域技术人员应当理解并不限于此。
当需要评分的照片为特写照片时，针对不能读取EXIF信息的照片，美学评价模型训练后得到单模型和多模型，美学评价模型通过机器学习训练得到。
如图7所示本发明用于对不能读取EXIF信息的特写照片进行评价的单模型训练流程图，用于对不能读取EXIF信息的特写照片进行评价的单模型训练按照如下方法进行训练：
S109、获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像。
S110、利用已知美学评分、图像场景类别标签和感兴趣区域位置坐标的若干个图像对预设的美学评价模型进行训练，得到单模型。
如图8所示本发明用于对不能读取EXIF信息的特写照片进行评价的多模型训练流程图，用于对不能读取EXIF信息的特写照片进行评价的多模型训练按照如下方法进行训练：
S111、获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像。
S112、利用已知美学评分和感兴趣区域位置坐标的若干个图像，根 据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与多个场景对应的多模型。实施例中示例性的以风景类图像照片和植物类图像照片为了进行多模型训练分别得到风景类图像美学评价的模型和植物类图像美学评价的模型，本领域技术人员应当理解并不限于此。
当需要评分的照片为特写照片时，针对能读取EXIF信息的照片，美学评价模型训练后得到单模型和多模型。
如图9所示本发明用于对能读取EXIF信息的特写照片进行评价的单模型训练流程图，用于对能读取EXIF信息的特写照片进行评价的单模型训按照如下方法进行训练：
S113、获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像。
S114、利用所述已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像对预设的美学评价模型进行训练，得到所述单模型。
如图10所示本发明用于对能读取EXIF信息的特写照片进行评价的多模型训练流程图，用于对能读取EXIF信息的特写照片进行评价的多模型训按照如下方法进行训练：
S115、获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像。
S116、利用已知美学评分、感兴趣区域位置坐标和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与多个场景对应的多模型。实施例中示例性的以风景类图像照片和植物类图像照片为了进行多模型训练分别得到风景类图像美学评价的模型和植物类图像美学评价的模型，本领域技术人员应当理解并不限于此。
下面实施例中具体说明本发明照片实现自动评分的过程：由上文中训练得到的不同种类的美学评价模型对照片进行自动评分。具体地，
对由智能设备中获取的照片采用图像场景分类算法对进行图像识别并进行场景分类，并判断照片是否为特写照片。本发明图像的来源可以 是已经存储的在计算设备中的照片，亦可获取智能设备摄像头实时取景图像作为图像来源。
采用图像场景分类算法对照片进行图像识别，根据识别结果将照片分为风景、夜景、建筑、动态、静态、逆光、人像、动物、植物多个类别。
实施例中，场景分类算法包括以下步骤实现照片的场景分类：
S201、采用物体识别算法识别图像中的物体。
S202、根据上下文语义模型分析出物体识别算法识别出的图像中物体之间的关联。
S203、根据分析出的结果可将图像归类为风景、夜景、建筑、动态、静态、逆光、人像、动物、植物多个类别，即得到图像所属场景分类结果。
应当理解对于场景分类算，本实施例中只是示例性的进行说明，本领域技术人员可以根据实际照片进行场景分类。
照片是否为特写照片通过如下方法判断：
通过显著性检测算法检测照片中的显著性区域；
当最大显著性区域面积/面积比例大于阈值时，则照片为特写照片，否则为非特写照片。
如图11所示不能读取EXIF信息的非特写照片的单模型美学评分流程图，当照片为不能读取EXIF信息的非特写照片时，则照片的第一信息包括图像场景分类结果和图像本身。将照片的第一信息输入到步骤S102得到的单模型进行评分，得到图像的美学评分。应当理解，对于输入单模型中的场景分类结果是图像场景不同类的类别标签。
如图12所示不能读取EXIF信息的非特写照片的多模型美学评分流程图，当照片为不能读取EXIF信息的非特写照片时，则照片的第一信息包括图像场景分类结果和图像本身。根据图像的场景分类结果，将不同类别场景的图像本身输入步骤S104得到多模型对应场景的美学评价模型进行评分，得到图像的美学评分。实施例中示例性的以风景类图像照片和植物类图像照片进行评分，分别得到风景类图像照片美学评分和植物 类图像照片美学评分，本领域技术人员应当理解并不限于此。
如图13所示图能读取EXIF信息的非特写照片的单模型美学评分流程图，当照片为不能读取EXIF信息的非特写照片时，则第一信息包括图像场景分类结果、图像照片EXIF、图像本身。将照片的第一信息输入到步骤S106得到的单模型进行评分，得到图像的美学评分。应当理解，对于输入单模型中的场景分类结果是图像场景不同类的类别标签。
如图14所示能读取EXIF信息的非特写照片的多模型美学评分流程图，当照片为能读取EXIF信息的非特写照片时，则照片的第一信息包括图像场景分类结果、图像照片EXIF和图像本身。根据图像的场景分类结果，将不同类别场景的图像照片EXIF和图像本身输入步骤S108得到多模型对应场景的美学评价模型进行评分，得到图像的美学评分。实施例中示例性的以风景类图像照片和植物类图像照片进行评分，分别得到风景类图像照片美学评分和植物类图像照片美学评分，本领域技术人员应当理解并不限于此。
本发明将照片进行场景分类后，判断照片为特写照片时，提取照片的感兴趣区域位置坐标，实施例中对于提取照片的感兴趣区域位置坐标通过显著性检测算法定位显著区域位置，将定位的显著区域位置作为感兴趣区域坐标。在一些实施例中，感兴趣区域位置坐标可以采用人工指定的区域位置作为感兴趣区域位置坐标。
如图15所示不能读取EXIF信息的特写照片的单模型美学评分流程图，当照片为不能读取EXIF信息的特写照片时，则第二信息包括图像场景分类结果、感兴趣区域位置坐标和图像本身。将照片的第二信息输入到步骤S110得到的单模型进行评分，得到图像的美学评分。应当理解，对于输入单模型中的场景分类结果是图像场景不同类的类别标签。
如图16所示不能读取EXIF信息的特写照片的多模型美学评分流程图，当照片为不能读取EXIF信息的特写照片时，则照片的第二信息包括图像场景分类结果、感兴趣区域位置坐标和图像本身。根据图像的场景分类结果，将不同类别场景的感兴趣区域位置坐标和图像本身输入步骤S112得到多模型对应场景的美学评价模型进行评分，得到图像的美学评分。实施例中示例性的以风景类图像照片和植物类图像照片进行评分， 分别得到风景类图像照片美学评分和植物类图像照片美学评分，本领域技术人员应当理解并不限于此。
如图17所示能读取EXIF信息的特写照片的单模型美学评分流程图，当照片为能读取EXIF信息的特写照片时，则第二信息包括图像场景分类结果、图像照片EXIF信息、感兴趣区域位置坐标和图像本身。将照片的第二信息输入到步骤S110得到的单模型进行评分，得到图像的美学评分。应当理解，对于输入单模型中的场景分类结果是图像场景不同类的类别标签。
如图18所示能读取EXIF信息的特写照片的多模型美学评分流程图，当照片为能读取EXIF信息的特写照片时，则照片的第二信息包括图像场景分类结果、图像照片EXIF信息、感兴趣区域位置坐标和图像本身。根据图像的场景分类结果，将不同类别场景的感兴趣区域位置坐标和图像本身输入步骤S114得到多模型对应场景的美学评价模型进行评分，得到图像的美学评分。实施例中示例性的以风景类图像照片和植物类图像照片进行评分，分别得到风景类图像照片美学评分和植物类图像照片美学评分，本领域技术人员应当理解并不限于此。
本发明一种基于美学评价的照片自动评分方法可以应用于照片自动分级、风景和人物拍摄引导、美学风格推荐。
本发明的基于美学评价的照片自动评分方法可以对智能设备中的相册照片的美感进行量化评分，通过对应用本发明的评分方法得到的评分进行评分范围划分，将评分结果分为若干等级，实现从美学角度对相册照片进行自动分级，并可按级别自动归类。
本发明的基于美学评价的图像照片自动评分方法可以对智能设备中实时取景的图像进行量化评分，有助于引导用户进行风景和人物等类照片的拍摄。
本发明的基于美学评价的图像照片自动评分方法，可以根据用户设备中的图像照片的评分结果，向用户推荐符合其美感的图像照片。
结合这里披露的本发明的说明和实践，本发明的其他实施例对于本领域技术人员都是易于想到和理解的。说明和实施例仅被认为是示例性的，本发明的真正范围和主旨均由权利要求所限定。

Claims (10)

  1. 一种基于美学评价的照片自动评分方法，其特征在于，所述方法包括在计算设备中执行的如下指令：
    训练美学评价模型；
    获取智能设备中的照片，采用图像场景分类算法对所述照片进行图像识别并进行场景分类，并判断所述照片是否为特写照片；
    当所述照片为非特写照片时，将所述照片的第一信息输入到所述美学评价模型中进行评分；
    当所述照片为特写照片时，将所述照片的第二信息输入到所述美学评价模型中进行评分。
  2. 根据权利要求1所述的方法，其特征在于，所述照片为非特写照片时，若所述照片不能读取EXIF信息，则所述第一信息包括图像场景分类结果和图像本身。
  3. 根据权利要求2所述的方法，其特征在于，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
    获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像；
    利用所述已知美学评分和图像场景类别标签的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
    利用已知美学评分的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。
  4. 根据权利要求1所述的方法，其特征在于，所述照片为非特写照片时，若所述照片可以读取EXIF信息，则所述第一信息包括图像场景分类结果、图像照片EXIF、图像本身。
  5. 根据权利要求4所述的方法，其特征在于，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
    获取已知美学评分、图像场景类别标签和EXIF信息的若干个图像；
    利用所述已知美学评分、图像场景类别标签和EXIF信息的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
    利用已知美学评分和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与场景对应的多模型。
  6. 根据权利要求1所述的方法，其特征在于，所述照片为特写照片时，提取所述照片的感兴趣区域位置坐标；若所述照片不能读取EXIF信息，则所述第二信息包括图像场景分类结果、感兴趣区域位置坐标、图像本身。
  7. 根据权利要求6所述的方法，其特征在于，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
    获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像；
    利用所述已知美学评分、图像场景类别标签和感兴趣区域位置坐标的若干个图像对预设的美学评价模型进行训练，得到所述单模型；
    利用已知美学评分和感兴趣区域位置坐标的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与多个场景对应的多模型。
  8. 根据权利要求1所述的方法，其特征在于，所述照片为特写照片时，提取所述照片的感兴趣区域位置坐标；若所述照片可以读取EXIF信息，则所述第二信息包括图像场景分类结果、图像照片EXIF、感兴趣区域位置坐标、图像本身。
  9. 根据权利要求8所述的方法，其特征在于，美学评价模型包括单模型和多模型，所述美学评价模型通过机器学习训练得到，其中所述美学评价模型的训练方法包括：
    获取已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像；
    利用所述已知美学评分、图像场景类别标签、感兴趣区域位置坐标和EXIF信息的若干个图像对预设的美学评价模型进行训练，得到所述单 模型；
    利用已知美学评分、感兴趣区域位置坐标和EXIF信息的若干个图像，根据图像所属的场景标签的不同，对预设的美学评价模型分别训练出与多个场景对应的多模型。
  10. 根据权利要求1所述的方法，其特征在于，所述照片是否为特写照片通过如下方法判断：
    通过显著性检测算法检测照片中的显著性区域；
    当最大显著性区域面积/面积比例大于阈值时，则所述照片为特写照片，否则为非特写照片。
PCT/CN2017/116059 2017-04-19 2017-12-14 一种基于美学评价的照片自动评分方法 WO2018192245A1 (zh)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN201710257193.2 2017-04-19
CN201710257193.2A CN107153838A (zh) 2017-04-19 2017-04-19 一种照片自动分级方法及装置
CN201710257194.7 2017-04-19
CN201710257194.7A CN107018330A (zh) 2017-04-19 2017-04-19 一种实时拍照指导方法及装置

Publications (1)

Publication Number Publication Date
WO2018192245A1 true WO2018192245A1 (zh) 2018-10-25

Family

ID=63855628

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/116059 WO2018192245A1 (zh) 2017-04-19 2017-12-14 一种基于美学评价的照片自动评分方法

Country Status (1)

Country Link
WO (1) WO2018192245A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111199540A (zh) * 2019-12-27 2020-05-26 Oppo广东移动通信有限公司 图像质量评价方法、装置、电子设备及存储介质
CN111382295A (zh) * 2018-12-27 2020-07-07 北京搜狗科技发展有限公司 一种图像搜索结果的排序方法和装置
CN112950580A (zh) * 2021-02-25 2021-06-11 北京金山云网络技术有限公司 质量评价方法、质量评价模型的训练方法和装置
CN113256134A (zh) * 2021-06-02 2021-08-13 岑英杰 一种基于美学工具箱的智能美学分析系统
CN111382295B (zh) * 2018-12-27 2024-04-30 北京搜狗科技发展有限公司 一种图像搜索结果的排序方法和装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102256030A (zh) * 2010-05-20 2011-11-23 Tcl集团股份有限公司 可匹配背景音乐的相册演示系统及其背景音乐匹配方法
CN104883548A (zh) * 2015-06-16 2015-09-02 金鹏电子信息机器有限公司 监控视频人脸抓取处理方法及其系统
CN107018330A (zh) * 2017-04-19 2017-08-04 中国电子科技集团公司电子科学研究院 一种实时拍照指导方法及装置
CN107153838A (zh) * 2017-04-19 2017-09-12 中国电子科技集团公司电子科学研究院 一种照片自动分级方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102256030A (zh) * 2010-05-20 2011-11-23 Tcl集团股份有限公司 可匹配背景音乐的相册演示系统及其背景音乐匹配方法
CN104883548A (zh) * 2015-06-16 2015-09-02 金鹏电子信息机器有限公司 监控视频人脸抓取处理方法及其系统
CN107018330A (zh) * 2017-04-19 2017-08-04 中国电子科技集团公司电子科学研究院 一种实时拍照指导方法及装置
CN107153838A (zh) * 2017-04-19 2017-09-12 中国电子科技集团公司电子科学研究院 一种照片自动分级方法及装置

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111382295A (zh) * 2018-12-27 2020-07-07 北京搜狗科技发展有限公司 一种图像搜索结果的排序方法和装置
CN111382295B (zh) * 2018-12-27 2024-04-30 北京搜狗科技发展有限公司 一种图像搜索结果的排序方法和装置
CN111199540A (zh) * 2019-12-27 2020-05-26 Oppo广东移动通信有限公司 图像质量评价方法、装置、电子设备及存储介质
CN112950580A (zh) * 2021-02-25 2021-06-11 北京金山云网络技术有限公司 质量评价方法、质量评价模型的训练方法和装置
CN113256134A (zh) * 2021-06-02 2021-08-13 岑英杰 一种基于美学工具箱的智能美学分析系统

Similar Documents

Publication Publication Date Title
CN109146892B (zh) 一种基于美学的图像裁剪方法及装置
Karayev et al. Recognizing image style
US9721183B2 (en) Intelligent determination of aesthetic preferences based on user history and properties
US8781175B2 (en) On-site composition and aesthetics feedback through exemplars for photographers
CN105740780B (zh) 人脸活体检测的方法和装置
US8649602B2 (en) Systems and methods for tagging photos
TW202006602A (zh) 三維臉部活體檢測方法、臉部認證識別方法及裝置
JP5524219B2 (ja) 対話式画像選択方法
CN103617432A (zh) 一种场景识别方法及装置
CN111583180B (zh) 一种图像的篡改识别方法、装置、计算机设备及存储介质
CN111160134A (zh) 一种以人为主体的视频景别分析方法和装置
WO2018192245A1 (zh) 一种基于美学评价的照片自动评分方法
CN106485186A (zh) 图像特征提取方法、装置、终端设备及系统
Dantone et al. Augmented faces
CN111401343B (zh) 识别图像中人的属性的方法、识别模型的训练方法和装置
WO2022127814A1 (zh) 一种图像的显著性物体检测方法、装置、设备及存储介质
Manh et al. Small object segmentation based on visual saliency in natural images
US8270731B2 (en) Image classification using range information
CN115410240A (zh) 智能人脸的痘痘和色斑分析方法、装置及存储介质
US10198791B2 (en) Automatic correction of facial sentiment of portrait images
JP6699048B2 (ja) 特徴選択装置、タグ関連領域抽出装置、方法、及びプログラム
WO2018192244A1 (zh) 一种智能设备的拍摄引导方法
CN116128530A (zh) 一种物品鉴定方法、系统、电子设备及可读存储介质
Wang et al. Aesthetic quality assessment for group photograph
CN111461248A (zh) 一种摄影构图线匹配方法、装置、设备及存储介质

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: 17906155

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17906155

Country of ref document: EP

Kind code of ref document: A1