Method and device for validating an image

Abstract

A method for validating a first image, implemented by a validation device comprising: - a step of calculating a conformity score for said first image by evaluating characteristics of the first image against a plurality of given image design rules, - if said conformity score is greater than a given threshold, a step of validating said first image. Figure for the abbreviation: Figure 1

Description

Title of the invention: Method and device for validating an image Scope of the invention

[0001] This invention relates to the field of telecommunications and in particular to image processing.

[0002] More specifically, the invention relates to a method for validating images. Previous art

[0003] Whether for artistic, commercial, educational or other purposes, the use of images has become central to informational communication systems. In particular, they play an important role in the design processes for video content, illustrated content, advertising content, video games, communication materials, etc.

[0004] However, before being used for specific purposes within these design processes, images are generally filtered or pre-selected to check that they comply with certain pre-established criteria or rules (format, image quality, permitted content, etc.).

[0005] Generally, reserves of images validated with regard to the criteria or rules in question are constituted by dedicated reference teams (for example texture managers, cartographers, brand / communication managers) and are made accessible to third parties wishing to use these images for particular purposes (graphic design, illustration, communication, etc.).

[0006] However, these validated image repositories are limited in size and do not always contain images corresponding to the specific needs and uses of a third-party user. In this case, the third-party user may use new images, which must be validated by a designated team before use, to ensure that these new images comply with the pre-established criteria or rules imposed upon them.

[0007] Image validation is generally based on their conformity to pre-established rules or charter(s) indicating all the criteria to be met for an image to be likely to be used for a given purpose.

[0008] These criteria may relate to technical elements of the image (e.g. resolution, size, colors used), the type of image (e.g. photo or drawing / pictogram), its composition (e.g. presence of a logo other than that of the company, position / size of the company logo, presence of a person or not) or the presence of text or not and, where applicable, the content / style of the text, etc.

[0009] However, this solution has the disadvantage of being relatively slow and time-consuming, as the images are analyzed using processes that may involve several levels of validation, which can prove problematic for entities or groups of people wishing to use said images quickly.

[0010] Indeed, this form of centralized organization of production and provision of images conforming to a charter is put under tension by the need of the different entities (and sub-entities) or organizations (divisions, services, project teams, ...) to be able to use the images quickly (for business needs, to illustrate or communicate, etc.).

[0011] To improve this situation, in some organizations, charters defining criteria for the conformity of images for a particular use are created and shared so that everyone can be able to verify the conformity of the images they wish to use.

[0012] This solution, however, also has drawbacks, as it requires training for those responsible for verifying the conformity of an image, which incurs costs. Furthermore, it is unpredictable in terms of results, since some conformity criteria can be complex or open to interpretation from one person to another, for example, the tone of a sentence, the "consensual" nature of an image, etc.

[0013] Moreover, the development of generative artificial intelligence introduces new problems in the field of image validation.

[0014] Generative artificial intelligences constitute a solution for rapidly producing images corresponding to an instruction (or request) formulated by a user (also referred to as a prompt in English), which makes it easy and quick to obtain a relevant image for a particular use (for example, to illustrate a text, to communicate about a product or service, etc.).

[0015] The increasing use of these tools nevertheless implies a multiplication of images that must be validated, indeed the design rules are not necessarily respected even if they are provided as input so that the conformity of the images is not acquired.

[0016] Object and summary of the invention

[0017] The invention notably remedies these drawbacks by proposing a method for validating a first image, implemented by a validation device and comprising: - a step of calculating a conformity score for said first image by evaluating characteristics of the first image with regard to a plurality of given image design rules, - if said conformity score is greater than a given threshold, a validation step of said first image.

[0018] This method advantageously allows for the automatic determination of whether an image conforms to specific design rules, for example, to a template or to technical prerequisites defined by said design rules. These design rules may relate in particular to various characteristics of the image such as its resolution, color palette, encoding, etc. Here, the term "given image design rules" refers to the set of rules, criteria, or points of conformity that an image must meet for its design to be validated, for example, for a given use (also referred to as its "intended use"). No limitations are attached to the use in question. For example, it may correspond to the creation of a texture catalog for the creation of three-dimensional images, or a thematic image bank (images of cars, images of cats, etc.).), pre-selection of images for compression processing or communication use.

[0019] In the following, the terms "criteria", "compliance criteria" and "graphic charter" are used interchangeably to refer to such design rules.

[0020] The solution also makes it possible to determine the conformity of said image more quickly and reliably compared to manual processing.

[0021] Since the process is automated, it is possible to process a large quantity of images in a homogeneous way, that is to say by applying the same image design rules to each image in a constant and uniform manner.

[0022] Beyond simply validating an image, obtaining a score also allows for comparing images based on their conformity score and thus determining which of two images best conforms to image design rules. Knowing such a score therefore makes it easy to select an image from among a plurality of images based on objective criteria, or to classify such images.

[0023] These rules can be used to frame the generation of an image or to evaluate whether an already created image can be used for particular purposes (online posting on a website, addition to a texture library, training of a device based on neural networks (MLP - Multilayer perceptron or Multilayer Perceptron in French), use in the context of corporate communication, school illustration etc.).

[0024] No limitation is attached to the nature of the design rules considered.

[0025] As mentioned previously, these may be, for example, rules relating to the technical characteristics of an image, such as its dimensions, the color ranges used, or to the composition of said image, for example the presence of a logo within the image and, where applicable, the nature and characteristics of said logo (such as the dimensions and position of the logo within the image).

[0026] Image design rules may also relate to the prohibited presence of certain specific elements within the image. These may include, for example, particular logos, the faces of specific personalities, or objects not authorized by the image design rules.

[0027] According to other examples, image design rules may also prohibit an image from not being royalty-free, an image from having already been used for specific purposes, or an image from having been retouched or generated by generative artificial intelligence.

[0028] Thus, the image design rules within the meaning of the invention may relate to characteristics concerning: - technical characteristics of the image: dimensions, nature or type of images, color ranges; - contextual elements relating to the design or use of the image: conditions of creation / generation, previous uses;

[0029] - more qualitative elements related to the content of the image: nature of the elements which factors include whether a message contained in an image uses correct or formal language, whether the image does not contain nudity, etc.

[0030] Of course this list is not exhaustive and other image design rules can be considered in a complementary or alternative way.

[0031] A score is calculated to reflect the conformity of the various image characteristics with the characteristics as defined by the image design rules. Thus, regardless of the type of rules considered, the score provides a quantitative and tangible measure of the image's conformity.

[0032] This score can be calculated via sub-scores where each sub-score corresponds to an assessment of conformity to a specific rule or criterion.

[0033] This may involve relating a quantitative value of a characteristic of the evaluated image to the expected value of that characteristic (for example, the size in megabytes of the image to the expected size).

[0034] In the case of a rule relating to qualitative values, other calculation methods can be implemented. For example, a compliance score relating to a design rule relating to "correct language" can be calculated based on the number of words identified as being part of a list of prohibited words; according to another example, nudity can be associated with a percentage of skin surface represented on the screen.

[0035] These are illustrative and non-limiting examples; the scores calculated in relation to compliance or non-compliance with design rules can be derived from any type of method known to a person skilled in the art: correspondence table, polynomial equations, logarithmics, etc.

[0036] According to a particular embodiment, if the conformity score is below the given threshold, the method includes a step of triggering a modification of the first image based on at least one said design rule not respected by the first image.

[0037] This embodiment advantageously allows for the rapid and targeted correction of an image to bring it into conformity with the evaluated design rules. A new image conforming to the image design rules can thus be efficiently obtained and used for specific purposes consistent with those design rules.

[0038] By integrating image modification to correct image elements that do not conform to certain image design rules, the invention makes it possible to reduce the processing time associated with creating new images from scratch that conform to the design rules.

[0039] Furthermore, this embodiment, which prioritizes correcting and modifying an image rather than generating a new one, makes it possible to obtain an image that retains its intrinsic qualities (i.e., its general characteristics, its content, the message conveyed, etc.). This is particularly important when the invention is used in contexts where the initial image (the first image) has been pre-selected based on these intrinsic qualities. This is the case, for example, when the first image is intended for communication, illustration, video editing, etc.

[0040] According to a particular embodiment, the step of triggering a modification of the first image includes sending a modification instruction to at least one image modification tool from among: - a color correction tool, - a text generation tool, - an image generation tool such as, for example, a generative artificial intelligence, - a missing part of image generation tool, - a logo insertion tool.

[0041] These examples are mentioned by way of illustration and are not limiting. This embodiment advantageously allows for the automatic triggering of a modification performed by a suitable tool based on the characteristic of the image that does not correspond or does not sufficiently correspond to a given image design rule and has resulted in an insufficient conformity score of the image (i.e., below the threshold considered).

[0042] Furthermore, in the specific context of generative artificial intelligences (AI), when the first image has been produced by a first intelligence In artificial generative intelligence, in response to generation instructions, the triggering step may include sending a second artificial generative intelligence an instruction to generate a second image comprising a noise field from the first image and said generation instructions.

[0043] Besides the fact that the images generated by these artificial intelligences do not necessarily conform to image design rules, the creation method tends to complicate their modification. Indeed, unlike traditional image production chains involving photographers, graphic designers / illustrators, and retouchers, users of generative artificial intelligences do not necessarily have the means and / or skills to edit and modify an image. If they do, they can only regenerate new images without any guarantee that these will conform to a given graphic charter.

[0044] This embodiment advantageously optimizes computing resources and processing times associated with modifying a first image when it has been generated by artificial intelligence.

[0045] Indeed, the return of the initial noise field (seed in English) and the generation prompt (prompt in English) makes it possible to limit the divergence between the first and second image and therefore to ensure that the modified image (i.e. the second image) is close to the initial image (i.e. first image) and retains its intrinsic qualities (content, message conveyed, etc.).

[0046] Limiting the divergence between the first and second images makes it possible, in particular, to avoid generating images that are too different from the initial image and therefore unsuitable for the intended use of the initial image. This solution has the effect of limiting the computing resources used to generate an image (computing power, electrical resources, etc.). The validation process according to the invention can then be applied to the second image, and so on.

[0047] According to this embodiment, the first generative artificial intelligence can be the same as the second or be a distinct artificial intelligence.

[0048] According to a particular embodiment, the conformity score is calculated from a weighted sum of individual scores associated respectively with each of the said design rules.

[0049] This embodiment advantageously allows for maximizing, or conversely minimizing, the importance of certain design rules relative to others. Indeed, while some rules are decisive and can lead to establishing that an image is non-compliant if they are not respected, other rules may be of lesser importance but allow for differentiation between two images in order to determine which one best conforms to the defined design rules.

[0050] Evaluating the image against given design rules makes it possible, where necessary, to determine the adjustments and modifications to be applied to the image to bring it into compliance with said rules. Thus, a user of the invention can, by accessing the individual score (i.e., the score specific to a characteristic or group of characteristics or more generally to a particular design rule), know the reason(s) why the image is not validated and the type of modification(s) to be made to bring the image into compliance.

[0051] This feedback also makes it possible to identify, where applicable, the type of action(s) to be taken to modify the image. This may involve contacting a person (editorial manager or author if the message is not compliant and needs to be regenerated, graphic designer, photographer) or triggering an action on a tool (colorimetry software, color matching software, homothetic correction software, etc.).

[0052] According to a particular embodiment, the weighting of the individual scores, associated with each of the said design rules, depends on a destination of the first image.

[0053] The destination of the first image refers to the type of use for which the image is intended.

[0054] As mentioned previously, the first image can be used to create a specific image collection such as a texture library, filtered for compression, used for communication purposes (for example, to create a library of content ready for use in brand communication, etc.). There are no limitations on the uses that can be envisaged.

[0055] These uses can also represent more specific destinations such as the creation of image backgrounds based on the content of the image or the type of subject represented (animals, landscape, etc.).

[0056] According to another example, the use of the image for specific communication purposes, for example for a physical display campaign (via posters or leaflets), for communication on an intranet site, in a presentation medium, in an animation or even in a message on a social network, all these forms of communication responding to different communication codes and incidentally to potentially different design rules.

[0057] This embodiment advantageously allows general image design rules to be adapted to take into account the specific characteristics related to particular uses and purposes of the image, and thus to determine the conformity of an image to these particular design rules. Advantageously, it is therefore possible to use a single general design rule and adapt it to the a multitude of existing communication forms rather than having to have design rules dedicated to each existing form of communication.

[0058] This embodiment is particularly advantageous since it allows the centralization of image design rules so that the images conform to a set of rules and criteria that are similar to each other, although adapted according to usage.

[0059] In a particular embodiment, the design rules relate to: - characteristics of a logotype identified on the first image (e.g. dimension, position on the image (in absolute or relative value), distortion compared to the identified reference logotype (respect for homothety, colorimetry)), - characteristics of a text present on said first image (e.g. syntactic, semantic and typographic), - the nature of the first image (e.g. pictorial or photographic, its stylistic classification), - descriptive characteristics of the first image (e.g. subjects and elements represented, author, date of creation, etc.), - technical characteristics of the first image (e.g. resolution, format, compression, vector or not, color palette), - the production conditions of the first image (e.g. image produced by a generative artificial intelligence, by a human).

[0060] In a particular embodiment, the process of validating a first image includes a step of generating at least one of the design rules from a framing document or an unstructured technical regulation, provided to the validation device.

[0061] By "unstructured technical guidelines or regulations document," we mean any document written in free language and intended to be consulted by a human. This could typically be a document presenting a company's graphic charter, a textual document outlining the language elements to be used or not used, or illustrative examples of images that do or do not conform to the design rules.

[0062] In general, such a document may contain: - numerical information referring for example to dimensions, RGB color codes or the maximum expected weight of an image, - more subjective information related to a requirement of "formal language", respect for "rules of decorum".

[0063] Moreover, this information is not necessarily presented in a structured way so that each piece of information is not necessarily explicitly associated with a particular design rule.

[0064] This embodiment advantageously allows the use of an unstructured framing document or technical regulations to define image design rules and thus automatically and simply configure the validation process of a first image implemented by the validation device.

[0065] The invention also relates to an image validation device comprising: - a calculation module configured to calculate a conformity score for said image by evaluating said image against a plurality of given image design rules, - a validation module configured to validate the image when said compliance score is greater than a given threshold.

[0066] The invention further relates to a computer program, comprising program code instructions for the implementation of a validation process according to any one of the particular embodiments described above, when this program is executed by a processor.

[0067] Such instructions can be stored permanently in a non-transient memory medium of a terminal implementing the image validation method according to the invention or, according to another embodiment, on a remote memory medium hosted on a third-party server.

[0068] This program may use any programming language, and be in the form of source code, object code, or intermediate code between source code and object code, such as in a partially compiled form, or in any other desirable form.

[0069] The invention also relates to a computer-readable information or recording medium on which a computer program as mentioned above is recorded.

[0070] The recording medium can be any entity or device capable of storing the program. For example, the medium can include a storage means, such as a ROM (Read Only Memory), for example a CD ROM (Compact Disc Read-Only Memory) or a microelectronic circuit ROM, or a magnetic recording means, for example a mobile medium, a hard disk or an SSD (Solid State-Drive).

[0071] On the other hand, the recording medium can be a transmissible medium, such as an electrical or optical signal, which can be transmitted via an electrical or optical cable, by radio, or by other means, so that the computer program it contains is executable remotely. The program according to the invention can, in particular, be uploaded to a network, for example, an Internet-type network.

[0072] Alternatively, the recording medium may be an integrated circuit in which one of the programs is incorporated, the circuit being adapted to execute or to be used in the execution of the validation process according to the invention.

[0073] The invention also relates to a system for validating a first image, said system comprising: - a device for validating the first image according to the invention, and - at least one computer tool for modifying images.

[0074] According to a particular embodiment of the first image validation system, the first image is produced by a first generative artificial intelligence in response to generation instructions and said at least one computer image modification tool includes a second generative artificial intelligence, the first image validation device being configured to send, to the second generative artificial intelligence, an instruction to generate a second image including a noise field of the first image and the generation instructions.

[0075] The validation device and system have the same advantages, mentioned above, as the validation process according to the invention.

[0076] It can also be envisaged, in other embodiments, that the process, the device, and the validation system according to the invention present in combination all or part of the aforementioned characteristics.

[0077] Brief description of the drawings and appendix

[0078] Other features and advantages will become apparent upon reading particular embodiments of the invention, given by way of illustrative and non-limiting examples, and the accompanying drawings, among which:

[0079] Figure [Fig. 1] represents, in its environment, a validation system according to the invention, in a particular embodiment,

[0080] Figure [Fig.2] schematically represents the hardware architecture of a device for validating a first image according to one embodiment of the invention,

[0081] Figure [Fig.3] represents the steps in the image validation process according to one embodiment,

[0082] Figure [Fig.4] represents an example of steps involved in triggering a modification of an image,

[0083] The appendix presents an example of a scoping document. Detailed description

[0084] Description of a system according to the invention, in a particular embodiment

[0085] Figure [Fig.1] represents, in its environment, a SYS system according to the invention in a particular embodiment, said system being configured to allow the validation of an image according to the invention.

[0086] To this end, the SYS system includes a device Tl for validating an image (first image within the meaning of the invention), in accordance with the invention. This can, for example, be a computer, a smartphone, or a server.

[0087] In the embodiment described here, the validation device Tl includes: - a Cal module configured to calculate a conformity score of the image to be validated by evaluating this image with regard to a plurality of given criteria associated with RC image design rules. - a Val module, configured to validate the image when the conformity score of said image is greater than a given threshold.

[0088] In the example of [Fig.1], the RC image design rules are defined by at least one Cad document received as input, grouping together a multitude of criteria to which an image must conform in order to be validated by the validation device T1.

[0089] In the example illustrated in [Fig.1], at least one first image Im is received as input in order to be validated by the validation device Tl, this image being unique or part of a group of images Glm comprising a multitude of other images.

[0090] In the embodiment described here, the SYS system further includes at least one Mod modification tool allowing an image to be modified according to at least one given criterion when the conformity score of this image is below the given threshold (it may be a resizing tool, colorimetry or any other image modification tool known to a person skilled in the art), as well as a first generative artificial intelligence IAG1 and a second generative artificial intelligence IAG2.

[0091] No limitation is attached to the nature of the Mod tool. The Mod tool may be a color correction tool, a text editing tool, an image retouching tool, an image generation tool, or any other image modification tool known to a person skilled in the art and which can be used to modify an image to make it conform to the RC design rules considered.

[0092] Generative artificial intelligences IAG1 and IAG2 are similar to generative artificial intelligences known to people skilled in the art. These may include LLMs (Large Language Models), based on neural network algorithms, Transformers, etc., such as the well-known LLMs MidJourney, Dalle e3, EleutherAI, etc.

[0093] In the embodiment described here, the generative artificial intelligences IAG1 and IAG2 are distinct. However, in an alternative embodiment, they can be considered as a single generative artificial intelligence.

[0094] In the embodiment described here, the various elements of the SYS validation system communicate via a network R. The network R can be a fiber optic, cellular, Wi-Fi network, or any type of network known to those skilled in the art. This network R is also used by the validation device Tl to access the Im image and the Cad document.

[0095] Alternatively, the Im image, where applicable the Glm image group, and the Cad document can be stored, in whole or in part, in a local memory of the validation device.

[0096] Fig. 2 presents the simplified structure of the validation device T1, configured to implement the validation process of a first image according to the invention.

[0097] In this particular embodiment of the invention, the validation device Tl has the classic hardware architecture of a computer and the steps executed by the device Tl, within the framework of the implementation of the method of validating an image of the present invention, as described later with reference to figure [Fig.3], are implemented by means of instructions of a computer program PG.

[0098] More specifically, the validation device Tl includes, in particular, a transmit / receive module ER, a memory MEM in which the computer program PG is stored, a processing unit UTR, equipped for example with a processor PROC, and controlled by the computer program PG stored in memory MEM. The memory MEM is a storage medium within the meaning of the invention.

[0099] The PG program defines functional modules of the Tl device which include in particular the calculation module Cal and the validation module Val introduced previously.

[0100] Description of the different stages of the image validation process according to an embodiment

[0101] Figure [Fig.3] describes the steps of the image validation process according to an embodiment in which these steps are implemented by the validation device Tl within the SYS system described in figure [Fig.l].

[0102] In E200, the validation device Tl receives an image Im to be validated (first image in the sense of the invention), for example for a given use. This image may or may not be from a group of images Glm to be validated. It may be sent manually to the validation device Tl by a user or sent via an automated application means integrated into another system not shown in [Fig. 1] (validation application, online publishing application integrating an image validation module, etc.).

[0103] There are no limitations attached to the nature of the image Im to be validated. It may be a 2D or 3D image, encoded in any format known to those skilled in the art (PNG, BMP, OBJ, X3D, MAX, etc.). It may also be a composite document including an image, for example a PDF or DOC file containing text and an image.

[0104] In E201, the validation device Tl calculates, using its calculation module Cal, a conformity score SC of the image Im by evaluating the image with regard to the image design rules RC, defined, in the embodiment described here, in the Cad document received as input and grouping a plurality of given criteria.

[0105] According to another embodiment, these design rules are directly pre-configured within the Cal calculation module of the validation device Tl.

[0106] To calculate the SC conformity score, the Tl validation device, and more specifically its Cal calculation module: - analyzes the image to be validated (and / or its metadata) to extract certain information (or characteristics) necessary to evaluate the various criteria defined by the image design rules. This information can be quantitative data (resolution, number of colors) or qualitative data (presence of text, logo, image style, etc.). - then, calculates the SC conformity score of the image using, in the embodiment described here, a calculation formula FC derived from the design rules RC, where the calculation formula allows the SC conformity score to be calculated from the values ​​of the information extracted during the analysis of the image.

[0107] Extracting information (or features) specific to the image may require the implementation of image analysis tools or tools for analyzing its metadata, which are known in themselves.

[0108] A calculation formula FC is derived from the design rules RC in the sense that these rules typically indicate the characteristics of a compliant image, i.e. having a maximum SC score ("what should be done") or a non-compliant image, i.e. having a minimum SC score ("what should not be done").

[0109] The FC calculation formula thus makes it possible, in the embodiment described here, to transcribe the design rules in the form of a mathematical or algorithmic formula using as variables the information (or characteristics) from the analysis of the image and relating to the different criteria defined by the image design rules.

[0110] Image design rules associate one or more image characteristics with expected values.

[0111] The table below gives non-limiting examples of design rules relating to different image characteristics. Image characteristics Examples of associated rules Image resolution 800x600, 1920x1080, ... Colors used Only colors belonging to a defined color palette (black; white; orange), included in: - RGB: 0,0,0; 256,256,256; 256,128,0; ... - CMYK: 0,0,0,1; 0,0,0,0; 0,70,90,0; ... Image type or "style": Photographic, drawing, infographic, flat design; Logo presence: Not allowed; Logo nature: Company logo only; Face(s) present: Yes, no; Text within the image: Yes, no, no more than 80 characters; Font used: Tahoma only, Arial or Arial Black only; Font size: 60px, 80em; Language type: Formal, professional, informal; Image acquisition process: Original image created by a human, image generated by generative intelligence; Previous uses and associated copyrights: Copyright, royalty-free, Creative Commons; ...

[0112] As mentioned above, the calculation of the SC conformity score, and upstream, the extraction and / or production of the image characteristics corresponding to each of the RC image design rules, can rely on the implementation of dedicated technical solutions, such as specific image analysis algorithms, such as pattern recognition, color detection (color filtering) and proportion comparison, OCR text recognition, etc.

[0113] A tool such as Google Lens or any other equivalent known to a person skilled in the art can be used to detect possible plagiarism or to verify that the image is correctly sourced (that the author indicated is indeed the creator of the image, etc.) in order to provide data relating to the design and uses of the image.

[0114] For face detection and facial landmark detection, algorithm libraries such as OpenCV (Haar cascades), SSD (Single Shot multiBox) are used. Detector), CNNs (ResNet-10), or deep learning algorithms (CNNs) can be used. These tools allow, in particular, the analysis of faces from any angle and can be used to identify people present in the image, thus enabling the calculation of a SC image compliance score that takes into account the presence or absence of people who should not appear in an image for a given use.

[0115] These examples are given only as examples, and other image analysis solutions may be considered as alternatives depending on the RC design rules to be evaluated.

[0116] Image design rules may vary depending on the intended use of the evaluated image. Therefore, the FC formula for calculating the SC conformity score will differ depending on the intended use.

[0117] For example, an image intended for printed communication may require CMYK (Cyan, Magenta, Yellow, Black) color encoding rather than RGB (Red, Green, Blue), a higher minimum resolution than an image intended for web graphics, etc. Thus, different image characteristics can be extracted and compared to specific criteria depending on the uses.

[0118] According to another example in which the validation process is used to determine which images are suitable for inclusion in a texture library, different image design rules can be applied to define "high definition" or "low definition" textures.

[0119] The FC calculation formula for calculating the SC conformity score may include the calculation of intermediate SCi scores relating to the evaluation of specific criteria defined by the design rules, such as those cited as examples above (resolution, colors used, presence of a face, etc.).

[0120] Just like the SC conformity score, the intermediate SCi scores can be recorded, in particular to be used to guide possible triggers of image modifications described in more detail later in step E203.

[0121] In the embodiment described here, a calculation formula FC is envisaged in which the conformity score SC corresponds to the average of intermediate scores SCi from the evaluation of the image characteristics with regard to the image design rules RC.

[0122] For example, if N denotes the number of criteria to be evaluated: SC = (SCi1 + SCi2 + SCi3 + ...) / N.

[0123] This is an illustrative and non-limiting example; a score can be calculated using other types of formulas, such as Boolean formulas assigning a specific SC score when a criterion is not met or assigning a specific SC score (for example 0) as soon as a single criterion is not met.

[0124] For example: if a colour used does not conform to the allowed colours, then the SC score = 0.

[0125] According to another example, the calculation of the SC compliance score may not involve an intermediate SCi score, relating to the different criteria taken individually, and take the form of a general formula.

[0126] For example SC = ([image width in pixels] x [image height in pixels]) / [number of colors used in the image].

[0127] The scores relating to each criterion can take any values ​​(discrete or within a predefined range), including negative values.

[0128] In addition, the validation device Tl can calculate the SC conformity score from a weighted sum of individual SCi scores assigned according to the different criteria considered.

[0129] For example: a compliance score can be calculated by applying a weighting coefficient of 10 to a score between 1 and -1 corresponding to the presence or absence of a logo other than that of a given company brand and a weighting coefficient of 1 to a score between 0 and 1 corresponding to the resolution of the image.

[0130] The values ​​of the weighting coefficients applied to the individual SCi scores may also vary depending on the type of use for which the image is intended. In such a case, a formula FC for calculating the SC conformity score can then be defined for each type of use.

[0131] For example: SC Usage: Institutional Communication SC = (Resolution criterion score) + 10 x (Logo criterion score) Internal Communication SC = (Resolution criterion score) + (Logo criterion score)

[0132] In the embodiment described here, the image design rules are transmitted to the TI validation device via the Cad reference document. This document also indicates that, according to the example, the design rules concerning the use of an image for institutional communication purposes are stricter with regard to the criterion related to the logo, hence a weighting of the score (Logo Criterion Score) within the FC calculation formula of the SC conformity score.

[0133] This Cad reference document can be formatted to be interpreted directly by the TL validation device. In this case, the document can take the form of a table listing, for example: - all the criteria to be taken into account, - the FC calculation formula (or method) of the SC compliance score and, where applicable, the calculation formula (or method) of each SCi sub-score, associated with each criterion individually taken into account in the (overall) SC compliance score.

[0134] For example, in the table below, the first line shows the FC calculation formula for the SC score and the method of calculating the different intermediate SCi scores that make it up. SC Note_hauteur_image x Note_largeur_image x Note_format Note_hauteur_hnage i ^(40 {[image height in pixels]) Note_largeur_image 1 ^(40 {[image width in pixels]) Note_format 1 / [image height in pixels] <r de l image en pixels]!

[0135] According to this example, an image that is not in "square" format with a resolution of 400 pixels by 400 pixels cannot obtain a score equal to 1.

[0136] According to another example, the Cad reference document may identify image analysis tools; these may be applications or functions known to the Tl device or, as in the table below, define functions that the Tl device must use to calculate the compliance score or scores associated with the different criteria.

[0137] For example: SC Color_OK() x Resolution_OK() Color_OK() = 1 if the image colors are contained in the list (#ff8000, #000000, #FFFFFF) = 1 - 0.1 x number of colors not contained in the list (#ff8000, #000000, #FFFFFF) Resolution_OK() = 1 for 4K resolution, 0.75 for 1080p resolution, 0.5 for SVGA resolution, 0.25 for resolutions lower than SVGA)

[0138] In the example given above, the detection of the colors used in the image or the determination of the image resolution are achieved through other image analysis tools (not described here). These tools can be written in any language known to a person skilled in the art (Java, JavaScript, C#, etc.).

[0139] Alternatively, the Cad reference document can be written in free, unstructured language and take the form of a graphic charter, a communication charter or a guide written for graphic designers or people (for example, employees of a company) who are required to communicate using images.

[0140] The Cad reference document can also be written in a semi-structured way, for example in the form of a table listing the different criteria constituting the image design rules and associating each criterion with a description written in free language.

[0141] An example of such a semi-structured CAD reference document is given below: Authorized Logo: No logos other than those on the Brand website. It is also essential that images do not contain any logos other than the company logo. Appropriate Logo Sizes: Two sizes: Master (600x600 pixels) and New Small (200x200 pixels). Authorized Fonts for Text: Typography should always be simple and clear, avoiding unnecessary details or embellishments. Our brand font is Helvetica Neue. Helvetica Neue 75 Bold is the standard user interface font used for company websites and is also used for headings, subheadings, and introductory text. Helvetica Neue 55 Roman is used for body text and tabular data.

[0142] Alternatively, the Cad reference document shown in the example above may include an additional column detailing the formulas for calculating intermediate scores.

[0143] By way of illustration, the appendix presents another example of a Cad reference document written in a semi-structured language.

[0144] This document may also contain illustrative examples of good and bad practices relating to the application or non-application of the expected conformity criteria.

[0145] In this case, the image validation device Tl may include an analysis module capable of extrapolating and explicating image design rules, mentioned in this document, in order to reformulate them into criteria interpretable and quantifiable by the calculation module Cal. Where appropriate, this analysis module also explicates a formula FC for calculating the SC conformity score.

[0146] According to another example, the validation device Tl can also receive a composite reference document comprising parts written in a structured manner, in a computer language directly interpretable by the validation device, and unstructured parts, such as examples of valid images and images that are not valid with regard to given image design rules.

[0147] Such an analysis module is typically a module implementing document analysis solutions such as multimodal language models, of the pre-trained generative transformer type, such as ChatGPT or Microsoft Copil or Gemini or any other solution known to the person skilled in the art capable of receiving design rules in an unstructured format and extracting a structured version allowing the configuration and provision of the various criteria of the validation device Tl of an image.

[0148] If the Cad reference document does not explicitly mention the rules for calculating the SC compliance score or the formulas for calculating the various intermediate SCi compliance scores associated with each criterion, the calculation module defines them itself according to predefined parameters.

[0149] For example, a Cad reference document consisting of free text stating "images should ideally be of 1080p resolution and should in no case include a logo other than that of the group" leads to the Tl device: - identifying the two criteria "resolution" and "unauthorized logo", - extracting information relating to resolution, identifying the presence or absence of a logo in the image and identifying it as an authorized logo, and - calculating the SC compliance score taking into account the difference in importance between the two criteria, for example by giving, within the generated FC calculation formula, a greater coefficient or weight to compliance with the logo criterion compared to that relating to resolution.

[0150] The CAD scoping document may also include RC design rules formulated as guidelines or instructions. In this case, the criterion is not explicitly stated, but provides information on the type of modification to be made to an image under certain conditions and indirectly on the criterion enabling the image to comply with the RC design rules. For example, the scoping document may include a rule such as "replace any logo identified on the image with the company logo."

[0151] According to this example, such a rule is interpreted as a prohibition of logos other than the one authorized (the company logo) and an indication of the type of modification to be triggered where appropriate in step E203, presented in more detail later.

[0152] It should be noted that the image design rules indicated in the Cad framing document can also be, in whole or in part, directly configured in the Tl validation device.

[0153] In E202, the Val module of the validation device Tl obtains the SC conformity score calculated by the Cal calculation module and validates or not the image.

[0154] According to the embodiment described herein, the Val module validates an image when the SC score is greater than a given VS threshold. The VS value can be predefined, entered in the Cad reference document, provided by a user of the validation device Tl, etc.

[0155] For example, the Cad reference document may take the following form: VS SC = 1: OK SC < 1: NOK SC Image_height x Image_width x Format_note Image_height i ^40 l-[image height in pixels]) Image_width 1 ^(401ÿimage width in pixels]) Format_note 1 {[image height in pixels]}

[0156] Optionally, the Val validation module can validate or not the image for several different types of use when several VS threshold values ​​are entered, each of these threshold values ​​being associated with a given use.

[0157] For example, uses can be associated with different VS thresholds, as in the following table presented for illustrative purposes, where each column is associated with a different use of the image: Image Use / Purpose: Institutional Communication, External Communication (Social Networks), Internal Communication. VS SC = 1: OK, SC < 1: NOK, SC > 0.75: OK, SC < 0.75: NOK, SC > 0.75: OK, SC < 0.75: NOK

[0158] According to a particular embodiment, when the type of use envisaged for the image Im is informed upstream of the calculation of the compliance score (for example institutional communication or communication on a social network), the device Tl compares only the compliance score SC calculated to the given threshold VS associated with this use.

[0159] According to a particular embodiment, the validation step may also involve comparing the intermediate compliance scores SCi to intermediate threshold values ​​VSi. This embodiment allows for more effective guidance of the image modification triggers described in step E203.

[0160] Optionally, the SC conformity score, (and where applicable the scores relating to specific criteria SCi), as well as the information that the image conforms to the RC image design rules, are sent to a third-party module, for example to the application or user who sent the Im image to the validation device Tl in order to verify its conformity.

[0161] In the embodiment described here, if said conformity score is less than the given threshold VS, the validation device Tl triggers during a step E203 a modification of the image Im by a module Mod according to at least one criterion not met by the image Im.

[0162] By way of example, an action involving a colorimetric modification can be triggered if the image Im does not conform to the colorimetric criteria defined in the design rules RC. In this case, the image is sent to a module Mod corresponding to or incorporating a colorimetric correction tool with an instruction to recolor the image from the allowed colors (using equivalence tables, tools based on a color quantization algorithm or any other solution known to the person skilled in the art).

[0163] Depending on the criterion not met, the triggering of an image modification may involve sending one or more modification instructions to the Mod tool(s), which may typically include: - a color correction tool, - a text generation tool, - an image generation tool, - a missing part of image generation tool, - a logo insertion tool, - a Routing tool, - a resizing tool, - an image compression tool.

[0164] These are tools cited by way of illustration and not limitation, step E203 of triggering an image modification based on at least one non-critical factor The image, as defined by the Im model, can utilize any image modification tool known to a person skilled in the art and adapted to the design rules in question. It should be noted that the same modification tool, Mod, can be used to perform multiple modifications (and thus integrate several of the tools listed above).

[0165] In the embodiment described here, the triggering of a modification and the sending of one or more instructions to specific tools depends on the intermediate SCi scores calculated in step E201, and their comparison to intermediate threshold values ​​VSi).

[0166] If several modifications to the image Im are necessary, the modifications can be made one after the other. Thus, a modified image (or intermediate image) can be returned to the image validation device Tl, which then sends it to another image modification application, or alternatively, an intermediate image is sent directly from a first modification application to a second modification application, and so on, until all the necessary modifications identified by the validation device Tl have been made.

[0167] Where appropriate, the sequence of the different modifications can be carried out in such a way as to limit the iterations of the calculation and modification steps according to instructions entered within the Tl device.

[0168] By way of illustration, a modification step resulting in the automatic generation / filling of an area of ​​the image (in order for example to hide / erase a logo) can be carried out after a color correction step, in order to take into account the possible introduction of new elements not conforming to the color references defined by the RC design rules.

[0169] Alternatively, these modifications can be carried out by external applications: - modifications to the content of the image can be carried out by applications such as Gimp or Photoshop, image addition / subtraction algorithms, merging of 2 images, image encoding / decoding, image transformation in a 2D plane, affine transformation, homography (via the use of Python OpenCV, numpy and imutils libraries), - Text modifications can be made by using LLM "text to text" to offer one or more alternatives to the detected text; -etc.

[0170] This type of modification requires several other modifications to be triggered: - to remove the original text from the image, - to add the new text - to fill in areas of the image not covered by the new text, - if necessary, to adjust the colorimetry of the image.

[0171] Figure [Fig.4] illustrates an example of steps implemented in triggering an image modification.

[0172] The image titled img_dst represents an image whose compliance score is below the expected threshold, notably due to the identification of a logo prohibited by the image design rules. Therefore, and within the context of this example, a logo change action is triggered.

[0173] A logo change tool is then implemented to adapt the image (img_src) of a reference logo in order to insert it as an overlay on the existing logo, which includes, for example, deformation of the image img_src and creation of a mask.

[0174] In the embodiment described here, the image Im evaluated by the validation device Tl is produced by a first generative artificial intelligence IAG1 in response to generation instructions (prompt).

[0175] When, during the validation step, the validation device Tl detects that at least one criterion of the image design rules SC is not met, the device Tl can trigger a modification of the image Im, including the regeneration of part or all of the image (for example, replacing a logo or removing a person's face if this is specified in the design rules, etc.). An instruction is then sent to a second generative artificial intelligence IAG2 to generate a second image. This instruction includes, in addition to the initial generation instructions for the image Im used by the first generative artificial intelligence IAG1, the initial noise field (seed) used to generate it.These elements (initial generation instructions and initial noise field) are assumed to be known and / or accessible by the device Tl; they can, for example, be received by the latter at the same time as the image Im to be evaluated or following a specific request.

[0176] Adding this initial noise field allows us to obtain a second image that is both close to the first image Im initially generated and conforms to the image design rules. This embodiment preserves many characteristics specific to the first image compared to a case where the image is regenerated from a different initial noise field (a seed) or according to a different instruction (a prompt).

[0177] It should be noted that the first and second generative artificial intelligences may be one and the same generative artificial intelligence, or distinct generative artificial intelligences.

[0178] Following the modification of the image Im by the second generative artificial intelligence, the modified image is sent here to the validation device Tl for validation and steps E201 and E202 (or even 203) are repeated.

[0179] In the example considered here, an image generated by an AI was taken into account, based on generation instructions and an initial noise field taken into account by the AI. However, the invention applies to any type of image, regardless of its origin or method of generation, such as photographs, illustrations produced by graphic designers, etc. The elements sent, if any, during step E203 (typically generation instructions and an initial noise field for an AI) naturally depend on the type of image considered.

[0180] In another embodiment, if the conformity score is below a given threshold (for example, the given VS threshold or another threshold defined for this purpose), the image is rejected. The E203 triggering step is therefore optional.

[0181] In another embodiment, the device Tl sends a third-party device a history of the compliance score calculated over the course of the iterations. This embodiment makes it possible, in particular, to verify that the image modification was useful and improved the compliance score. It is therefore a way to evaluate the effectiveness of the image modification module(s) triggered by the image validation device.

[0182] Annex Criteria Expected Value Minimum Resolution Minimum 200x200 pixels Maximum Size No more than 10MB Logo Allowed Logo No logo other than the Brand website logo. Also, it is essential that the images do not contain any logos other than the company logo. Appropriate Logo Sizes 2 sizes: Master (600x600 pixels) and New Small (200x200 pixels). Logo Placement in the Image Master Logo in publication headers, welcome pages, New Small for digital applications, smartphones Logo Color A single brand color ##79001 Allowed Fonts for Text Typography should always be simple and clear, avoiding unnecessary details or embellishments. Our brand font is Helvetica Neue. Ivetica Neue 75 Bold is the standard user interface font used for the company's websites and is also Helvetica Neue 55 Roman is used for titles, subtitles, and introductory text. Helvetica Neue 55 Roman is used for the body text and tabular data. Text Formatting Rules: Text Alignment: Typography should always be left-aligned. This ensures consistent content anchoring and a uniform layout. Letter and Line Spacing: Line lengths should be managed to represent the correct appearance of the text. Limit the range of line lengths. Shorter lines conform to the brand and are more legible. 20 to 40 characters per line for short lines of text or short paragraphs. 60 to 80 characters per line for the main text. More than 80 characters is too long. Font Restrictions: Only brand fonts are accepted.Brand Color Palette: Color Codes to Use (CMYK, RGB, Pantone) The brand uses the following colors: black, white, and orange. Orange should be used as a highlight or focus color. Permitted Color Combinations: Use of the Brand Color. To ensure that the brand color meets accessibility standards when used in combination with other base colors, two shades have been chosen to represent the brand on screens: "brand color" and "accessible color." Brand Color: This color is used as the digital version of the brand color. The brand color, when used in combination with black, meets AAA accessibility standards (8:1 contrast ratio). Accessible Color: This shade has been chosen to provide sufficient minimum contrast with white to meet accessibility standards. Compliance with AA accessibility standards when using text (3:1 contrast ratio). Color usage guidelines for text, background, and graphic elements: Primary Colors: These are the core colors that represent the brand's visual identity. In this case, the primary colors are black, white, and orange. Supporting Colors: These are colors used to complement the primary colors and add visual variety. They can be used for design elements such as backgrounds, borders, etc. Functional Colors: These are colors used for interactive or functional elements, such as buttons, links, icons, etc. They can be used to draw the user's attention and indicate possible actions.Functional grays: These are shades of gray used for elements such as backgrounds, dividers, muted text boxes, etc. They are used to provide visual hierarchy and readability. Color balance: This refers to the balanced use of different colors in the design to create visual harmony and a pleasant user experience. This may include the use of contrasting colors, complementary colors, or consistent color schemes. The style and aesthetic of the images that reflect the Ea brand identity must always support the idea of ​​communication as well as our personality: approachable, simple, positive, and bold. Light, colors, and saturation must be natural; we no longer add supporting colors to our photographs. Our photos are taken up close and express a spontaneous moment in life.The guidelines concerning the brand include guidelines on images but also on the use of images with people or animations. ills Restrictions on facial recognition in images: Human faces must not be recognizable. Tone: Correct or formal language. "AI" Marking: Where applicable, adding a statement indicating that the image was created by AI is mandatory. This statement must be customized according to the brand specifications. The specific statement should be "Designed by us, generated by AI."

Claims

Demands

1. Method for validating a first image (Im), implemented by a validation device (Tl) and comprising: - a calculation step (E201) of a conformity score (SC) of said first image by evaluating characteristics of the first image with regard to a plurality of given image design rules (RC), - if said conformity score is greater than a given threshold (VS), a validation step (E202) of said first image.

2. Method for validating a first image (Im) according to claim 1 comprising, if said conformity score (SC) is less than said given threshold (VS), a triggering step (E203) of a modification of said first image (Im) according to at least one said design rule (RC) not respected by said first image.

3. Method for validating a first image (Im) according to claim 2 wherein the triggering step (E203) of a modification of said first image (Im) comprises sending a modification instruction to at least one image modification tool (Mod) from among: - a color correction tool, - a text generation tool, - an image generation tool, - a missing part of image generation tool, - a logo insertion tool.

4. A method for validating a first image (Im) according to claim 2 or 3 wherein said first image was produced by a first generative artificial intelligence (IAG1) in response to generation instructions and wherein the step of triggering a modification of said first image includes sending to a second generative artificial intelligence (IAG2) an instruction to generate a second image comprising a noise field of the first image and said generation instructions.

5. A method for validating a first image (Im) according to any one of the preceding claims, wherein the score of Compliance (SC) is calculated from a weighted sum of individual scores (SCi) associated with each of the said design rules.

6. Method for validating a first image (Im) according to claim 5 wherein a weighting in said weighted sum of individual scores (SCi) associated with each of said design rules (RC) depends on a destination of said first image.

7. Method for validating a first image (Im) according to any one of the preceding claims wherein said design rules (RC) relate to: - technical characteristics of a logo identified on said first image, - characteristics of a text present on said first image, - chromatic characteristics of said first image, - a nature of said first image, - descriptive characteristics of said first image, - technical characteristics of said first image, - its production conditions of said first image.

8. Method for validating a first image (Im) according to any one of the preceding claims comprising a step of generating at least one of said design rules (RC) from a framing document (Cad) or unstructured technical regulation, provided to the validation device.

9. Image validation device (Tl) comprising: - a calculation module (Cal) configured to calculate a conformity score (CS) of said image by evaluating said image against a plurality of given image design rules (RC), - a validation module (Val) configured to validate said image when said conformity score is greater than a given threshold.

10. System (SYS) for validating a first image, said system comprising: - a device for validating the first image (Im) according to claim 9, and - at least one computer tool for modifying images.

11. System (SYS) for validating a first image (Im) according to claim 10, wherein said first image is produced by a first generative artificial intelligence (IAG1) in response to generation instructions and wherein said at least one computer image modification tool comprises a second generative artificial intelligence (IAG1), the first image validation device being configured to send, to said second generative artificial intelligence, an instruction to generate a second image comprising a noise field of the first image and said generation instructions.