WO2021031677A1 - 一种目标对象的banner图的批量自动生成方法及装置 - Google Patents
一种目标对象的banner图的批量自动生成方法及装置 Download PDFInfo
- Publication number
- WO2021031677A1 WO2021031677A1 PCT/CN2020/096991 CN2020096991W WO2021031677A1 WO 2021031677 A1 WO2021031677 A1 WO 2021031677A1 CN 2020096991 W CN2020096991 W CN 2020096991W WO 2021031677 A1 WO2021031677 A1 WO 2021031677A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- target
- target object
- master
- layer
- banner
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
Definitions
- the present invention relates to the technical field of image processing, in particular to a method and device for automatically generating a banner map of a target object in batches.
- the embodiments of the present invention provide a method and device for automatically generating banner graphs of target objects in batches, so as to overcome the long manual design cycle, large workload, high repetitiveness and materials in the prior art. Can not be fully utilized and other issues.
- the technical solution adopted by the present invention is:
- a method for batch automatic generation of banner graphs of target objects includes the following steps:
- processing the batch of target object pictures to obtain the target object body and target object picture information corresponding to each target object picture includes:
- the matching at least two target masters corresponding to each target object body from a master library according to the target object body and the target object picture information includes:
- the layer information of each target master is obtained, the elements corresponding to each layer of each target master are obtained according to the layer information, and the grayscale processed according to each target master Layer, determining the color matching of the copywriting corresponding to the layer includes:
- the method further includes:
- the method further includes:
- the banner graph output with the score satisfying the preset condition is filtered from the batch of banner graphs.
- a device for automatically generating banner graphs of target objects in batches includes:
- the image processing module is used to process batches of target object pictures, and obtain the target object body and target object picture information corresponding to each target object picture;
- a master matching module configured to match at least two target masters corresponding to each target object body from a master library according to the target object body and the target object picture information;
- the material obtaining module is used to obtain the layer information of each target master, obtain the elements corresponding to each layer of each target master according to the layer information, and after graying processing according to each target master To determine the color matching of the copywriting corresponding to the layer;
- the banner image generation module is used to generate batches of information based on the target object body, the target object picture information, the at least two target masters, the elements corresponding to each layer, and the copywriting and the color matching of the copywriting banner illustration.
- the image processing module includes:
- the image segmentation unit is used to segment each target object picture by using an image segmentation algorithm
- the image optimization unit is used for smoothing the obtained segmentation results, and obtaining the target object body corresponding to each target object picture;
- the information extraction unit is configured to obtain target object picture information corresponding to each target object picture according to each target object picture and the target object body.
- the master matching module includes:
- the distance calculation unit is configured to calculate the distance between each target object body and each master in the master library according to the target object picture information
- the master selection unit is used to select at least two masters with the closest distance to each of the target objects as target masters.
- the material acquisition module includes:
- the master parsing unit is configured to perform structural analysis on the at least two target masters respectively, and obtain layer information corresponding to each target master;
- the element matching unit is used to match elements corresponding to each layer of each target master from the material pool according to the layer information corresponding to each target master;
- the copywriting color matching unit is used to perform grayscale processing on the layer of each target master, and obtain the color matching of the copywriting corresponding to the layer according to the color of the grayscale processed layer.
- the device further includes:
- the size judgment module is used to judge whether the size of the at least two target masters is consistent with the target size
- the size expansion module is used to expand the size of the at least two target masters so that the size of the at least two target masters is consistent with the target size.
- the device further includes:
- the banner image scoring module is used to extract the target features in each banner image of the batch of banner images, and calculate the score of each banner image according to all the target features;
- the banner image screening module is used to filter out the banner images whose scores meet preset conditions from the batch of banner images according to the scores of the banner images.
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention, by matching multiple masters from a pre-maintained master library according to the batch of target object pictures uploaded by users, and determining layers according to the masters Corresponding elements and copy color matching, and then generate banner images in batches according to batch target object pictures, multiple masters, and elements and copy color matching corresponding to layers, shorten the design cycle, avoid repetitive work, and improve the production efficiency of banner images;
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention perform segmentation processing on batches of target object pictures through an image segmentation algorithm, and then smooth the segmentation results to obtain the target object body and improve the target object The quality of the main image, thereby improving the quality of the final generated banner image;
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention expand the size of the target master to make the size of the target master consistent with the size required by the user, thereby achieving only A few masters need to be maintained to cover all sizes and improve the utilization of masters.
- Fig. 1 is a flow chart showing a method for automatically generating banner images of target objects in batches according to an exemplary embodiment
- Fig. 2 is a flow chart of processing a batch of target object pictures to obtain the target object body and target object picture information corresponding to each target object picture according to an exemplary embodiment
- Fig. 3 is a diagram showing obtaining the layer information of each target master according to an exemplary embodiment, obtaining the elements corresponding to each layer of each target master according to the layer information, and according to each target master The layer after gray-scale processing, the flow chart of determining the color matching of the copywriting corresponding to the layer;
- Fig. 4 is a schematic structural diagram of an apparatus for automatically generating banner images of target objects in batches according to an exemplary embodiment.
- Fig. 1 is a flowchart of a method for batch automatic generation of banner images of target objects according to an exemplary embodiment. Referring to Fig. 1, the method includes the following steps:
- S1 Process the batch of target object pictures, and obtain the target object body and target object picture information corresponding to each target object picture.
- the target subject includes commodities and the like.
- the subject of the target object is the object to be prominently expressed by the banner image.
- a high-quality target object image ie, the target object subject
- the target image information includes the color, category, suitable scene, style, size and other information of the target object.
- preprocessing is also required for the batch of target object pictures, and the preprocessing includes at least the batches uploaded by users.
- the quality of the target image is judged, and blurry images with low resolution are filtered out.
- S2 Match at least two target masters corresponding to each target object body from a master library according to the target object body and the target object picture information.
- a master library is maintained in advance in the embodiment of the present invention.
- the masters in the master library are maintained in accordance with certain label dimensions, such as size, style, category, color, scene, layout, etc., that is, the masters of different scenes, sizes, and layouts are maintained in the master library. Version.
- each banner image corresponds to a master. Because it is necessary to generate batches of banner images for users to choose from, and to ensure the diversity of the generated banner images, in the embodiment of the present invention, it is necessary to match each of them from the master library according to the target object subject and target object picture information. At least two target masters corresponding to the target object body.
- the banner map synthesized by each target master represents a series, which not only allows a certain aesthetic reference when generating the banner map, but also ensures the richness of the banner map.
- S3 Obtain the layer information of each target master, obtain the elements corresponding to each layer of each target master according to the layer information, and obtain the gray-scaled layer according to each target master, Determine the color scheme of the copy corresponding to the layer.
- each master in the master library contains regularly named structured layers.
- the master in the embodiment of the present invention is set to contain multiple layers, such as the target object body layer and the copywriting. Layers, background layers, etc. It should be noted here that, in the embodiment of the present invention, the number of layers in the master is not limited, and the user can set it according to actual needs.
- the layer information of each target master is obtained, and then the elements corresponding to each layer of each target master are obtained according to the layer information, and the gray-scale processing is performed according to each target master After the layer, determine the color of the copy corresponding to the layer.
- S4 Generate a batch of banner images according to the target object body, the target object picture information, the at least two target masters, the elements corresponding to each layer, and the color matching of the copywriting and copywriting.
- the target object body, the target object picture information, the target master, the elements corresponding to each layer, the copywriting and the color matching of the copywriting and other materials obtained in the above steps are synthesized to generate a batch of banner images.
- the action of generating batches of banner graphs is abstracted into a feature matching process.
- the layers in the selected target master are abstracted into different dimensions, such as color, size, theme, shape, texture, space, etc., and quantified as a feature template.
- the element corresponding to each layer is matched from the material library, the feature distance between the element and the template is calculated to determine whether it can be matched, and the banner map is generated according to the matching result.
- the specific matching algorithm is not limited, and the user can select or set according to actual needs.
- Fig. 2 is a flow chart showing processing batches of target object pictures according to an exemplary embodiment to obtain the target object subject and target object picture information corresponding to each target object picture.
- the processing of batches of target object pictures to obtain the target object subject and target object picture information corresponding to each target object picture includes:
- S101 Perform segmentation processing on each target object picture by using an image segmentation algorithm, perform smoothing processing on the obtained segmentation result, and obtain a target object body corresponding to each target object picture.
- an image segmentation algorithm based on deep learning is used to segment each picture in the batch of target object pictures, that is, the target object body and the background image are preliminarily segmented, and the target object body is cut out from the target object picture. Then, an anti-aliasing algorithm is used to smooth the edges of the segmentation result (that is, the target object body obtained by the preliminary segmentation) to output a high-quality target object body.
- the output result may be a transparent image of the subject of the target object.
- S102 Acquire target object picture information corresponding to each target object picture according to each of the target object pictures and the target object body.
- the target image information of each target image can be extracted through a convolutional neural network (CNN).
- the target image information includes the color, category, suitable scene, style, size and other information of the target object.
- the at least two objects corresponding to each target object are matched from a master library according to the target object body and the target object picture information.
- the target master includes:
- the closest master is retrieved by Euclidean distance, and this is used as the design manuscript of the banner map.
- the masters obtained by each query are not unique.
- the banner graph synthesized by each target master represents a series, so that even when the banner graph is designed There is a certain aesthetic reference to ensure the richness of the banner map.
- Fig. 3 is a diagram showing obtaining the layer information of each target master according to an exemplary embodiment, obtaining the elements corresponding to each layer of each target master according to the layer information, and according to each target master After graying the layer, the flow chart for determining the color matching of the copywriting corresponding to the layer is shown in FIG. 3.
- the acquisition of each target The layer information of the master, the element corresponding to each layer of each target master is obtained according to the layer information, and the layer corresponding to each target master is determined according to the gray-scaled layer of each target master
- the color scheme of the copy includes:
- S301 Perform structural analysis on the at least two target masters respectively, and obtain layer information corresponding to each target master.
- each target master is structured and analyzed to obtain the layer information corresponding to each target master.
- the layer information includes background, logo, copywriting, tile decoration, product area decoration, copy area decoration, fragment decoration, welt decoration, wire frame decoration and other information.
- S302 Match the elements corresponding to each layer of each target master from the material pool according to the layer information corresponding to each target master.
- a material pool is constructed in advance according to different layer information of the master, and elements corresponding to each layer are collected in the material pool.
- tags are maintained through multi-classification algorithms, such as aspect ratio, color, style, shape, texture, space, size, applicable scenarios, etc.
- a single element can correspond to multiple dimensions of tag data.
- the layer information corresponding to each target master is parsed according to the above steps, and the elements corresponding to each layer of each target master are filtered from the material pool according to the matching rule algorithm.
- the specific matching rule algorithm is not restricted, and the user can select or set according to actual needs.
- the background image is also maintained in the material pool in advance.
- the maximum width and height of the banner map required by the business side does not exceed 1246.
- all background images are maintained at a size of 1300 ⁇ 1300, and the background images are first compressed to the longest side of the size required by the user during synthesis. , And then crop to the user's required size based on the center point of the compressed square map background. This ensures that only compression and cropping operations are performed during synthesis, and the picture is prevented from being stretched and distorted, thereby ensuring the quality of the combined picture.
- the size of the above background image is only an exemplary description, and does not constitute a limitation to the embodiment of the present invention. In actual application, the user can maintain the background image in the material pool according to actual needs.
- S303 Perform grayscale processing on the layer of each target master, and obtain the color matching of the copywriting corresponding to the layer according to the color of the grayscale processed layer.
- the copy of the banner image can be uploaded by the user.
- Judge the color of the copy by the color of the layer where the copy is located.
- each layer of the target master is grayed out, and then the color depth is judged according to a certain threshold, and the color of the copy is judged according to the color depth, such as white or black, etc., to avoid low contrast and difficult text Recognition.
- the method also includes:
- the masters in the master library it is impossible for the masters in the master library to cover all user input sizes.
- the size required by the user ie, the user The size of the input banner map
- expand the master to the size required by the user according to the principle that the position of the element corresponding to the layer remains unchanged relative to the banner map. That is, the topological relationship between the local relative position and the global absolute position of the corresponding element of the layer in the banner map is used as the objective function to expand, and the expansion range is 50% of the aspect ratio.
- the method further includes:
- the banner graph output with the score satisfying the preset condition is filtered from the batch of banner graphs.
- the banner evaluation system mainly evaluates the banner graph from two aspects: scoring items and scoring dimensions.
- the target features extracted from the banner graph are used as scoring items to score the banner graph.
- the scoring item is the element item that the banner needs to be scored and marked.
- the main scoring items are: whether the color matching is coordinated, whether the layout is reasonable, whether the style is uniform, whether the product map is coordinated, etc.
- Scoring dimension that is, to score the generated banner from different dimensions. Integrating the banner from design to its online performance and other factors, in the embodiment of the present invention, the scoring dimensions are summarized as the designer, the online exposure and click ratio of the banner, and ordinary offline users.
- a large number of design products are used as training data to train and generate a banner evaluation system.
- the scoring is mainly determined by operators and designers to determine the score label, and input into the network for training output results.
- the corresponding banner score label is determined by the online exposure click ratio of the output result, the banner image is displayed according to the score, and finally fed back to the model for continuous reinforcement learning to ensure that the evaluation mechanism is continuously improved from multiple dimensions.
- a preset condition is also set.
- a banner graph whose score meets the preset condition is output to the user for selection.
- Fig. 4 is a schematic structural diagram of an apparatus for automatically generating banner images of target objects in batches according to an exemplary embodiment.
- the apparatus includes:
- the image processing module is used to process batches of target object pictures, and obtain the target object body and target object picture information corresponding to each target object picture;
- a master matching module configured to match at least two target masters corresponding to each target object body from a master library according to the target object body and the target object picture information;
- the material obtaining module is used to obtain the layer information of each target master, obtain the elements corresponding to each layer of each target master according to the layer information, and after graying processing according to each target master To determine the color matching of the copywriting corresponding to the layer;
- the banner image generation module is used to generate batches of information based on the target object body, the target object picture information, the at least two target masters, the elements corresponding to each layer, and the copywriting and the color matching of the copywriting banner illustration.
- the image processing module includes:
- the image segmentation unit is used to segment each target object picture by using an image segmentation algorithm
- the image optimization unit is used for smoothing the obtained segmentation results, and obtaining the target object body corresponding to each target object picture;
- the information extraction unit is configured to obtain target object picture information corresponding to each target object picture according to each target object picture and the target object body.
- the master matching module includes:
- the distance calculation unit is configured to calculate the distance between each target object body and each master in the master library according to the target object picture information
- the master selection unit is used to select at least two masters with the closest distance to each of the target objects as target masters.
- the material acquisition module includes:
- the master parsing unit is configured to perform structural analysis on the at least two target masters, respectively, to obtain layer information corresponding to each target master;
- the element matching unit is used to match elements corresponding to each layer of each target master from the material pool according to the layer information corresponding to each target master;
- the copywriting color matching unit is used to perform grayscale processing on the layer of each target master, and obtain the color matching of the copywriting corresponding to the layer according to the color of the grayscale processed layer.
- the device further includes:
- the size judgment module is used to judge whether the size of the at least two target masters is consistent with the target size
- the size expansion module is used to expand the size of the at least two target masters so that the size of the at least two target masters is consistent with the target size.
- the device further includes:
- the banner image scoring module is used to extract the target features in each banner image of the batch of banner images, and calculate the score of each banner image according to all the target features;
- the banner image screening module is configured to filter out the banner images whose scores meet preset conditions from the batch of banner images according to the scores of the banner images.
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention, by matching multiple masters from a pre-maintained master library according to the batch of target object pictures uploaded by users, and determining layers according to the masters Corresponding elements and copy color matching, and then generate banner images in batches according to batch target object pictures, multiple masters, and elements and copy color matching corresponding to layers, shorten the design cycle, avoid repetitive work, and improve the production efficiency of banner images;
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention perform segmentation processing on batches of target object pictures through an image segmentation algorithm, and then smooth the segmentation results to obtain the target object body and improve the target object The quality of the main image, thereby improving the quality of the final generated banner image;
- the method and device for batch automatic generation of banner graphs of target objects provided by the embodiments of the present invention expand the size of the target master to make the size of the target master consistent with the size required by the user, thereby achieving only A few masters need to be maintained to cover all sizes and improve the utilization of masters.
- the apparatus for automatically generating banner images of target objects in batches triggers the banner image generation business
- only the division of the above functional modules is used as an example for illustration. In actual applications, the above can be changed according to needs. Function allocation is completed by different functional modules, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.
- the device for automatically generating banner images of target objects in batches provided in the above-mentioned embodiments belongs to the same idea as the method for automatically generating banner images of target objects in batches, that is, the device is a method for automatically generating banner images of target objects in batches. Yes, please refer to the method embodiment for the specific implementation process, which will not be repeated here.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Information Transfer Between Computers (AREA)
Abstract
一种目标对象的banner图的批量自动生成方法及装置，该方法包括：对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息（S1）；根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版（S2）；获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色（S3）；根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图（S4）。通过根据目标对象图片自动批量生成banner图，缩短设计周期，避免重复性工作，提高banner图制作效率。
Description
本发明涉及图像处理技术领域，特别涉及一种目标对象的banner图的批量自动生成方法及装置。
随着互联网技术的普及和发展，网上购物成为越来越多人生活中密不可分的一部分。各大电商平台通过对商品图片进行装饰，并加上促销文字的宣传信息，来吸引用户的关注和点击，提高商品交易转化率。在此环节中，生成一张优质的商品banner图就显得至关重要。
在图像处理技术尚未运用到视觉设计领域之前，banner图的生成主要是由设计师来完成的。在此过程中，设计师们往往会因为一些简单的需求付出相当多的时间，如文案内容的修改，坑位的多尺寸拓展等，而这些大量的重复性的工作对设计师的进步成长起到的作用却非常有限。与此同时，精准营销是电商未来发展的趋势，在大流量背景下，首页的海报资源展位需要展示“千人千面”的效果(其中，“千人千面”是一种推荐算法，其依托购物网站巨大的数据库，构建出买家的爱好模型，它能从细分类目中抓取那些特征与买家爱好点匹配的推行产品，展示在目标客户阅读的网页上，帮助卖家确定潜在买家，完成精准营销)，这对海报的生产效率也提出了非常高的要求。
综上所述，亟需提出一种新的banner图的生成方法以满足上述需求。
发明内容
为了解决现有技术的问题，本发明实施例提供了一种目标对象的banner图 的批量自动生成方法及装置，以克服现有技术中，人工设计周期长、工作量大、重复性高以及素材不能被充分利用等问题。
为解决上述一个或多个技术问题，本发明采用的技术方案是：
一方面，提供了一种目标对象的banner图的批量自动生成方法，该方法包括如下步骤：
对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息；
根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版；
获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色；
根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图。
进一步的，所述对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息包括：
利用图像分割算法对每一目标对象图片进行分割处理，对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体；
根据所述每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
进一步的，所述根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版包括：
根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离，选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
进一步的，所述获取每一目标母版的图层信息，根据所述图层信息获取每 一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色包括：
对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息；
根据每一目标母版对应的所述图层信息从素材池中匹配出每一目标母版的每个图层对应的元素；
将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
进一步的，根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版后，所述方法还包括：
判断所述至少两个目标母版的尺寸是否与目标尺寸一致，若不一致，则对所述至少两个目标母版进行尺寸拓展，使所述至少两个目标母版的尺寸与所述目标尺寸一致。
进一步的，所述方法还包括：
提取所述批量的banner图中每一banner图中的目标特征，根据所有所述目标特征计算每一banner图的分数；
根据所述banner图的分数从所述批量的banner图中筛选出分数满足预设条件的banner图输出。
另一方面，提供了一种目标对象的banner图的批量自动生成装置，所述装置包括：
图像处理模块，用于对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息；
母版匹配模块，用于根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版；
素材获取模块，用于获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后 的图层，确定所述图层对应的文案的配色；
banner图生成模块，用于根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图。
进一步的，所述图像处理模块包括：
图像分割单元，用于利用图像分割算法对每一目标对象图片进行分割处理；
图像优化单元，用于对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体；
信息提取单元，用于根据每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
进一步的，所述母版匹配模块包括：
距离计算单元，用于根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离；
母版选取单元，用于选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
进一步的，所述素材获取模块包括：
母版解析单元，用于对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息；
元素匹配单元，用于根据每一目标母版对应的所述图层信息从素材池中匹配出每一目标母版的每个图层对应的元素；
文案配色单元，用于将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
进一步的，所述装置还包括：
尺寸判断模块，用于判断所述至少两个目标母版的尺寸是否与目标尺寸一致；
尺寸拓展模块，用于对所述至少两个目标母版进行尺寸拓展，使所述至少 两个目标母版的尺寸与所述目标尺寸一致。
进一步的，所述装置还包括：
banner图评分模块，用于提取所述批量的banner图中每一banner图中的目标特征，根据所有所述目标特征计算每一banner图的分数；
banner图筛选模块，用于根据所述banner图的分数从所述批量的banner图中筛选出分数满足预设条件的banner图输出。
本发明实施例提供的技术方案带来的有益效果是：
1、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置，通过根据用户上传的批量目标对象图片从预先维护的母版库中匹配多个母版，根据母版确定图层对应的元素及文案配色，然后根据批量目标对象图片、多个母版以及图层对应的元素及文案配色批量生成banner图，缩短设计周期，避免重复性工作，从而提高banner图的制作效率；
2、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置，通过图像分割算法对批量目标对象图片进行分割处理，再对分割结果进行平滑处理，获取目标对象主体，提高目标对象主体图像的质量，从而提高最终生成的banner图的质量；
3、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置，通过对目标母版进行尺寸拓展使目标母版的尺寸与用户需求的尺寸一致，从而实现在母版库中仅需维护少量母版即可覆盖所有尺寸，提高母版的利用率。
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是根据一示例性实施例示出的目标对象的banner图的批量自动生成方 法的流程图；
图2是根据一示例性实施例示出的对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息的流程图；
图3是根据一示例性实施例示出的获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色的流程图；
图4是根据一示例性实施例示出的目标对象的banner图的批量自动生成装置的结构示意图。
为使本发明的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
图1是根据一示例性实施例示出的目标对象的banner图的批量自动生成方法的流程图，参照图1所示，该方法包括如下步骤：
S1：对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息。
具体的，本发明实施例中，目标对象主体包括商品等。目标对象主体是banner图所要突出表达的对象，生成一张优质的banner图首先需要一张高质量的目标对象图(即目标对象主体)。因此，需要先对用户上传的批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息。其中目标对象图片信息包括目标对象主体的颜色、类目、适合场景、风格、尺寸等信息。
这里需要说明的是，为了进一步提高banner图的质量，本发明实施例中， 在对批量目标对象图片进行处理前，还需对批量目标对象图片进行预处理，预处理至少包括对用户上传的批量目标对象图片进行质量判断，过滤掉模糊分辨率不高的图片。
S2：根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版。
具体的，为了使得生成的banner图尽可能的符合设计结构美学，本发明实施例中预先维护有母版库。母版库中的母版按照一定的标签维度进行维护，如尺寸、风格、类目、颜色、场景、布局方式等，即在母版库中维护了不同场景、尺寸和布局方式等维度的母版。在合成banner图时，每张banner图都对应了一个母版。由于需要生成批量的banner图供用户选择，同时为了保证生成的banner图的多样性，因此，在本发明实施例中需要根据目标对象主体以及目标对象图片信息从母版库中匹配出与每一目标对象主体对应的至少两个目标母版。每个目标母版合成的banner图代表一个系列，这样既使得banner图生成时有一定的美学参考，又保证了banner图的丰富性。
S3：获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色。
具体的，母版库中每个母版均包含规则化命名的结构化图层，作为一种示例，设置本发明实施例中的母版包含多个图层，如目标对象主体图层、文案图层、背景图层等。这里需要说明的是，本发明实施例中，不对母版中图层的数量进行限制，用户可以根据实际需求进行设置。在匹配出目标母版后，获取每个目标母版的图层信息，然后根据图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定图层对应的文案的配色。
S4：根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner 图。
具体的，将上述步骤获取到的目标对象主体、目标对象图片信息、目标母版、每个图层对应的元素以及文案及文案的配色等素材进行合成，生成批量的banner图。作为一种较优的实施方式，本发明实施例中，将生成批量的banner图的这一动作抽象成特征匹配的过程。具体实施时，在匹配算法的特征提取阶段，将选定的目标母板中的图层抽象化出不同的维度，如颜色、尺寸、主题、形状、纹理、空间等，量化后作为特征模板。在特征匹配阶段，从素材库中匹配出的每个图层对应的元素，计算元素与模板之间的特征距离决定能否匹配，根据匹配结果生成banner图。这里需要说明的是，本发明实施例中，不对具体的匹配算法进行限定，用户可以根据实际需求进行选取或设置。
图2是根据一示例性实施例示出的对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息的流程图，参照图2所示，作为一种较优的实施方式，本发明实施例中，所述对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息包括：
S101：利用图像分割算法对每一目标对象图片进行分割处理，对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体。
具体的，利用基于深度学习的图像分割算法对批量目标对象图片中的每一张图片进行分割处理，即将目标对象主体与背景图进行初步分割，将目标对象主体从目标对象图片中抠出。然后利用抗锯齿算法对分割结果(即初步分割得到的目标对象主体)的边缘进行平滑处理，输出高质量的目标对象主体。作为一种示例，输出结果可以是目标对象主体的透明图。
S102：根据所述每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
具体的，可以通过卷积神经网络(CNN)提取出每一张目标对象图片的目标对象图片信息，目标对象图片信息包括目标对象主体的颜色、类目、适合场 景、风格、尺寸等信息。
作为一种较优的实施方式，本发明实施例中，所述根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版包括：
根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离，选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
具体的，本发明实施例中，通过欧式距离检索查询距离最近的母版，以此作为banner图的设计原稿。需要注意的是每次查询得到的母版并不是唯一的，通过设置距离阈值，可以筛选得到多个目标母版，每种目标母版合成的banner图代表一个系列，这样既使得banner图设计时有一定的美学参考，又保证了banner图的丰富性。
图3是根据一示例性实施例示出的获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色的流程图，参照图3所示，作为一种较优的实施方式，本发明实施例中，所述获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色包括：
S301：对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息。
具体的，匹配出目标母版后，分别对每个目标母版进行结构化解析，获取每个目标母版对应的图层信息。图层信息包括背景、logo、文案、平铺修饰、商品区域修饰、文案区域修饰、碎片装饰、贴边装饰、线框装饰等信息。
S302：根据每一目标母版对应的所述图层信息从素材池中匹配出每一目标母版的每个图层对应的元素。
具体的，本发明实施例中，预先根据母版的不同图层信息构建一个素材池， 素材池中收集有与每个图层对应的元素。元素上传至素材池时通过多分类算法维护标签，如高宽比、颜色、风格、形状、纹理、空间、尺寸、适用场景等，单个元素可对应多个维度标签数据。根据上述步骤解析获取的每个目标母版对应的图层信息，按照匹配规则算法从素材池中筛选出每个目标母版的每个图层对应的元素。这里需要说明的是，本发明实施例中，不对具体的匹配规则算法进行限制，用户可以根据实际需求进行选取或设置。
作为一种较优的实施方式，本发明实施例中，素材池还预先维护了背景图。通过调研统计业务方的banner尺寸，发现业务方所需的banner图最大宽高尺寸不超过1246。而为了适应业务的不同尺寸需求，本发明实施例中，在素材池中，所有背景图都按1300×1300尺寸进行维护，在合成时首先将背景图等比压缩到用户所需尺寸最长边，然后基于压缩后的方图背景中心点裁剪到用户的需求尺寸。这样保证了合成时只做压缩和裁剪操作，避免图片被拉伸失真，从而保证合图质量。这里需要说明的是，上述背景图的尺寸只是一种示例性的说明，并不对本发明实施例构成限制，实际应用时，用户可以根据实际需求维护素材池中的背景图。
S303：将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
具体的，banner图的文案可以由用户上传。通过文案所在图层的颜色的深浅来判断文案的配色。具体实施时，将目标母版的每个图层进行灰度化处理，然后按一定的阈值判断颜色深浅，根据颜色深浅判断文案的配色，如取白色还是黑色等，避免对比度过低导致文本难以识别。
作为一种较优的实施方式，本发明实施例中，根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版后，所述方法还包括：
判断所述至少两个目标母版的尺寸是否与目标尺寸一致，若不一致，则对所述至少两个目标母版进行尺寸拓展，使所述至少两个目标母版的尺寸与所述 目标尺寸一致。
具体的，由于用户需求的多样性，母版库中的母版不可能覆盖掉所有的用户输入尺寸，当用户输入的尺寸与查询得到的母版不相符时，根据用户需求的尺寸(即用户输入的banner图的尺寸)，按照图层对应的元素相对banner图位置不变的原则对母版进行拓展，拓展到用户需求的尺寸。即以图层对应元素在banner图中的局部相对位置与全局绝对位置的拓扑关系作为目标函数进行拓展，拓展范围为高宽比50％，通过此方法母版库中仅需要维护少量母版即可覆盖所有尺寸，保证了母版的多样性。在实际应用中，同种布局母版只需维护四种不同比例，即可覆盖站内所有尺寸。
作为一种较优的实施方式，本发明实施例中，所述方法还包括：
提取所述批量的banner图中每一banner图中的目标特征，根据所有所述目标特征计算每一banner图的分数；
根据所述banner图的分数从所述批量的banner图中筛选出分数满足预设条件的banner图输出。
具体的，为了使呈现给用户进行选取的banner更为优质，本发明实施例中，还可以设置将生成的banner图片输入banner评价系统进行打分，根据分值(即打分结果)对批量的banner图进行排序，根据排序结果筛选出批量的最终banner图，将最终banner图呈现给用户进行选择。banner评价系统主要从打分项和打分维度两方面评价banner图，其中本发明实施例中，将从banner图中的提取的目标特征作为打分项对banner图进行打分。打分项，顾名思义即为该banner需要被打分标记的元素项。目前主要的打分项有：配色是否协调、布局是否合理、风格是否统一、商品图是否协调等。打分维度：即从不同的维度对生成的banner进行打分。综合该banner从设计到其线上表现等因素，本发明实施例中，把打分维度归纳为设计师、banner的线上曝光点击比和普通的线下用户。
将大量的设计成品作为训练数据训练生成banner评价系统，具体实施时，在banner评估模型训练阶段，打分主要由运营人员和设计师来决定得分标签， 输入到网络中进行训练输出结果。在模型验证阶段，由输出结果的线上曝光点击比例来决定对应的banner的得分标签，按照得分展示banner图片，最终再反馈回模型不断的强化学习，保证从多个维度不断完善该评价机制。
这里需要说明的是，本发明实施例中，还设置有预设条件，当从批量banner图中筛选出分数满足预设条件的banner图输出给用户进行选择。
图4是根据一示例性实施例示出的目标对象的banner图的批量自动生成装置的结构示意图，参照图4所示，该装置包括：
图像处理模块，用于对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息；
母版匹配模块，用于根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版；
素材获取模块，用于获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色；
banner图生成模块，用于根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图。
作为一种较优的实施方式，本发明实施例中，所述图像处理模块包括：
图像分割单元，用于利用图像分割算法对每一目标对象图片进行分割处理；
图像优化单元，用于对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体；
信息提取单元，用于根据每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
作为一种较优的实施方式，本发明实施例中，所述母版匹配模块包括：
距离计算单元，用于根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离；
母版选取单元，用于选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
作为一种较优的实施方式，本发明实施例中，所述素材获取模块包括：
母版解析单元，用于对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息；
元素匹配单元，用于根据每一目标母版对应的所述图层信息从素材池中匹配出每一目标母版的每个图层对应的元素；
文案配色单元，用于将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
作为一种较优的实施方式，本发明实施例中，所述装置还包括：
尺寸判断模块，用于判断所述至少两个目标母版的尺寸是否与目标尺寸一致；
尺寸拓展模块，用于对所述至少两个目标母版进行尺寸拓展，使所述至少两个目标母版的尺寸与所述目标尺寸一致。
作为一种较优的实施方式，本发明实施例中，所述装置还包括：
banner图评分模块，用于提取所述批量的banner图中每一banner图中的目标特征，根据所有所述目标特征计算每一banner图的分数；
banner图筛选模块，用于根据所述banner图的分数从所述批量的banner图中筛选出分数满足预设条件的banner图输出。
综上所述，本发明实施例提供的技术方案带来的有益效果是：
1、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置，通过根据用户上传的批量目标对象图片从预先维护的母版库中匹配多个母版，根据母版确定图层对应的元素及文案配色，然后根据批量目标对象图片、多个母版以及图层对应的元素及文案配色批量生成banner图，缩短设计周期，避免重复性工作，从而提高banner图的制作效率；
2、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置， 通过图像分割算法对批量目标对象图片进行分割处理，再对分割结果进行平滑处理，获取目标对象主体，提高目标对象主体图像的质量，从而提高最终生成的banner图的质量；
3、本发明实施例提供的目标对象的banner图的批量自动生成方法及装置，通过对目标母版进行尺寸拓展使目标母版的尺寸与用户需求的尺寸一致，从而实现在母版库中仅需维护少量母版即可覆盖所有尺寸，提高母版的利用率。
需要说明的是：上述实施例提供的目标对象的banner图的批量自动生成装置在触发banner图生成业务时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将装置的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的目标对象的banner图的批量自动生成装置与目标对象的banner图的批量自动生成方法实施例属于同一构思，即该装置是基于该目标对象的banner图的批量自动生成方法的，其具体实现过程详见方法实施例，这里不再赘述。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。
以上所述仅为本发明的较佳实施例，并不用以限制本发明，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 (10)
- 一种目标对象的banner图的批量自动生成方法，其特征在于，所述方法包括如下步骤：对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息；根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版；获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色；根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图。
- 根据权利要求1所述的目标对象的banner图的批量自动生成方法，其特征在于，所述对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息包括：利用图像分割算法对每一目标对象图片进行分割处理，对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体；根据所述每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
- 根据权利要求1或2所述的目标对象的banner图的批量自动生成方法，其特征在于，所述根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版包括：根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离，选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
- 根据权利要求1或2所述的目标对象的banner图的批量自动生成方法，其特征在于，所述获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色包括：对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息；根据每一目标母版对应的所述图层信息从素材池中匹配出每一目标母版的每个图层对应的元素；将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
- 根据权利要求1或2所述的目标对象的banner图的批量自动生成方法，其特征在于，根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版后，所述方法还包括：判断所述至少两个目标母版的尺寸是否与目标尺寸一致，若不一致，则对所述至少两个目标母版进行尺寸拓展，使所述至少两个目标母版的尺寸与所述目标尺寸一致。
- 根据权利要求1或2所述的目标对象的banner图的批量自动生成方法，其特征在于，所述方法还包括：提取所述批量的banner图中每一banner图中的目标特征，根据所有所述目标特征计算每一banner图的分数；根据所述banner图的分数从所述批量的banner图中筛选出分数满足预设条件的banner图输出。
- 一种目标对象的banner图的批量自动生成装置，其特征在于，所述装置包括：图像处理模块，用于对批量目标对象图片进行处理，获取每一目标对象图片对应的目标对象主体以及目标对象图片信息；母版匹配模块，用于根据所述目标对象主体以及所述目标对象图片信息从母版库中匹配出与每一所述目标对象主体对应的至少两个目标母版；素材获取模块，用于获取每一目标母版的图层信息，根据所述图层信息获取每一目标母版的每个图层对应的元素，以及根据每一目标母版灰度化处理后的图层，确定所述图层对应的文案的配色；banner图生成模块，用于根据所述目标对象主体、所述目标对象图片信息、所述至少两个目标母版、所述每个图层对应的元素以及所述文案及文案的配色生成批量的banner图。
- 根据权利要求7所述的目标对象的banner图的批量自动生成装置，其特征在于，所述图像处理模块包括：图像分割单元，用于利用图像分割算法对每一目标对象图片进行分割处理；图像优化单元，用于对获取到的分割结果进行平滑处理，获取每一目标对象图片对应的目标对象主体；信息提取单元，用于根据每一目标对象图片以及所述目标对象主体获取每一目标对象图片对应的目标对象图片信息。
- 根据权利要求7或8所述的目标对象的banner图的批量自动生成装置，其特征在于，所述母版匹配模块包括：距离计算单元，用于根据所述目标对象图片信息计算每一所述目标对象主体与母版库中各个母版的距离；母版选取单元，用于选取至少两个与每一所述目标对象主体距离最近的母版作为目标母版。
- 根据权利要求6或7所述的目标对象的banner图的批量自动生成装置，其特征在于，所述素材获取模块包括：母版解析单元，用于对所述至少两个目标母版分别进行结构化解析，获取每一目标母版对应的图层信息；元素匹配单元，用于根据每一目标母版对应的所述图层信息从素材池中匹 配出每一目标母版的每个图层对应的元素；文案配色单元，用于将所述每一目标母版的图层进行灰度化处理，根据灰度化处理后的图层的颜色获取所述图层对应的文案的配色。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CA3148915A CA3148915A1 (en) | 2019-08-21 | 2020-06-19 | Method and device for automatically generating banner images of a target object in batches |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910777228.4A CN110659371B (zh) | 2019-08-21 | 2019-08-21 | 一种目标对象的banner图的批量自动生成方法及装置 |
| CN201910777228.4 | 2019-08-21 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021031677A1 true WO2021031677A1 (zh) | 2021-02-25 |
Family
ID=69037701
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2020/096991 WO2021031677A1 (zh) | 2019-08-21 | 2020-06-19 | 一种目标对象的banner图的批量自动生成方法及装置 |
Country Status (3)
| Country | Link |
|---|---|
| CN (1) | CN110659371B (zh) |
| CA (1) | CA3148915A1 (zh) |
| WO (1) | WO2021031677A1 (zh) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659371B (zh) * | 2019-08-21 | 2022-07-01 | 苏宁云计算有限公司 | 一种目标对象的banner图的批量自动生成方法及装置 |
| CN111353532A (zh) * | 2020-02-26 | 2020-06-30 | 北京三快在线科技有限公司 | 图像生成方法及装置、计算机可读存储介质、电子设备 |
| CN113450433B (zh) * | 2020-03-26 | 2024-08-16 | 阿里巴巴集团控股有限公司 | 图片生成方法、装置、计算机设备和介质 |
| CN113628105A (zh) * | 2020-05-07 | 2021-11-09 | 阿里巴巴集团控股有限公司 | 图像处理方法、装置、存储介质和处理器 |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140365310A1 (en) * | 2013-06-05 | 2014-12-11 | Machine Perception Technologies, Inc. | Presentation of materials based on low level feature analysis |
| CN107909631A (zh) * | 2017-11-29 | 2018-04-13 | 商派软件有限公司 | 一种图片合成方法 |
| CN108345624A (zh) * | 2017-01-24 | 2018-07-31 | 阿里巴巴集团控股有限公司 | 生成页面的方法及装置 |
| CN110659371A (zh) * | 2019-08-21 | 2020-01-07 | 苏宁云计算有限公司 | 一种目标对象的banner图的批量自动生成方法及装置 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102124446A (zh) * | 2008-08-08 | 2011-07-13 | 汤姆逊许可证公司 | 动态横幅组合方法 |
| US11270350B2 (en) * | 2012-11-30 | 2022-03-08 | 3I Avi, Llc | Systems and method for verifying vehicle banner production and image alteration |
| CN107025676B (zh) * | 2016-01-25 | 2021-02-02 | 阿里巴巴集团控股有限公司 | 一种图片模板以及图片的生成方法及相关装置 |
| CN108961362A (zh) * | 2017-05-27 | 2018-12-07 | 阿里巴巴集团控股有限公司 | 一种网络图片的生成方法与装置 |
| CN109308729B (zh) * | 2017-07-27 | 2023-01-24 | 阿里巴巴集团控股有限公司 | 图片合成处理方法、装置及系统 |
| CN109961493A (zh) * | 2017-12-26 | 2019-07-02 | 苏宁云商集团股份有限公司 | 展示页面上的横幅广告图片生成方法及装置 |
| CN109741425B (zh) * | 2019-01-04 | 2023-07-25 | 广州方硅信息技术有限公司 | 横幅图片生成方法、装置及存储介质、计算机设备 |
-
2019
- 2019-08-21 CN CN201910777228.4A patent/CN110659371B/zh active Active
-
2020
- 2020-06-19 WO PCT/CN2020/096991 patent/WO2021031677A1/zh active Application Filing
- 2020-06-19 CA CA3148915A patent/CA3148915A1/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140365310A1 (en) * | 2013-06-05 | 2014-12-11 | Machine Perception Technologies, Inc. | Presentation of materials based on low level feature analysis |
| CN108345624A (zh) * | 2017-01-24 | 2018-07-31 | 阿里巴巴集团控股有限公司 | 生成页面的方法及装置 |
| CN107909631A (zh) * | 2017-11-29 | 2018-04-13 | 商派软件有限公司 | 一种图片合成方法 |
| CN110659371A (zh) * | 2019-08-21 | 2020-01-07 | 苏宁云计算有限公司 | 一种目标对象的banner图的批量自动生成方法及装置 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110659371A (zh) | 2020-01-07 |
| CA3148915A1 (en) | 2021-02-25 |
| CN110659371B (zh) | 2022-07-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021031677A1 (zh) | 一种目标对象的banner图的批量自动生成方法及装置 | |
| US11282116B2 (en) | Image quality assessment to merchandise an item | |
| CN110889883A (zh) | 一种自适应的智能横幅广告图片生成方法及系统 | |
| Chen et al. | DGCA: high resolution image inpainting via DR-GAN and contextual attention | |
| CN109308729B (zh) | 图片合成处理方法、装置及系统 | |
| CN110750666A (zh) | 图片生成方法、系统、电子设备及存储介质 | |
| Ahmadi et al. | Context-aware saliency detection for image retargeting using convolutional neural networks | |
| CN117058271A (zh) | 用于生成商品主图背景的方法及计算设备 | |
| CN110232726B (zh) | 创意素材的生成方法及装置 | |
| US20220375223A1 (en) | Information generation method and apparatus | |
| EP4195136B1 (en) | Automated video generation from images for e-commerce applications | |
| CN111243061A (zh) | 一种商品图片的生成方法、装置、系统 | |
| CN117372570A (zh) | 广告图像生成方法、装置 | |
| US20230129431A1 (en) | One-to-Many Automatic Content Generation | |
| You et al. | Automatic synthesis of advertising images according to a specified style | |
| He et al. | Wordart designer: User-driven artistic typography synthesis using large language models | |
| CN111158647B (zh) | 基于结构化理论的创意素材自适应生成方法及装置 | |
| CN112837332A (zh) | 创意设计的生成方法、装置、终端、存储介质及处理器 | |
| JP2020502710A (ja) | ウェブページメイン画像認識方法及び装置 | |
| CN113869960B (zh) | 海报生成方法、装置、存储介质及计算机设备 | |
| WO2022089427A1 (zh) | 视频生成方法、装置、电子设备以及计算机可读介质 | |
| Guangpeng et al. | Computer vision technology based on image optical processing in visual packaging art design simulation | |
| TW202143173A (zh) | 智能影像拼接系統及拼接方法 | |
| CN117392276A (zh) | 图像处理方法、设备及存储介质 | |
| Shao | Exploration of the Integration and Innovation of Digital Art Elements and Traditional Cultural and Creative Designs |
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: 20854143 Country of ref document: EP Kind code of ref document: A1 |
|
| ENP | Entry into the national phase |
Ref document number: 3148915 Country of ref document: CA |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20854143 Country of ref document: EP Kind code of ref document: A1 |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20854143 Country of ref document: EP Kind code of ref document: A1 |