Method for determining the color of facial skin and corresponding system

Abstract

A method for determining the skin color of a user's area of interest comprises: - importing from an image library (LIB) of a user computing device (APP) at least one image (IMk) comprising a representation of the user's area of interest; - performing a treatment (TS) of the at least one image (IMk) with a machine learning model (AI_TS) adapted to provide, for each imported image (IMk), a numerical value (ClrEstm_k) representative of the skin color of the area of interest present in the at least one imported image (IMk); - performing an evaluation (EVAL) of the skin color of the area of interest based on the one or more numerical values (ClrEstm_k) of each imported image.

Description

[0001] The embodiments and examples of the present invention relate to, for example but not limited to, a method for determining facial skin color and a corresponding system, particularly in the context of recommending cosmetic products.

[0002] Online product purchases (especially via mobile computing devices) are a practical way for consumers to browse and obtain products. Various technologies exist to analyze online buyers' browsing history, purchase history, reviews, product ratings, and other data to provide product recommendations that online buyers may be interested in.

[0003] However, for certain types of products, face-to-face interaction is difficult to replace with online experiences. For example, beauty products such as foundation or other cosmetics are difficult to search for online and also difficult to recommend automatically. This is primarily due to the fact that it is difficult to automatically estimate the consumer characteristics (i.e., in particular facial skin color) that are the most relevant basis for recommendations. Even if images or videos of consumers are captured, they are generally not reliable for determining skin color due to the limitations of image processing technology in uncontrolled environments. Inconsistent lighting conditions at different locations will result in different colors being reproduced in images or videos from different locations; therefore, the determined skin color will vary depending on the lighting conditions.

[0004] Patent application PCT / US 2020 / 041350, filed on July 9, 2020 and claiming priority to provisional patent application US 16 / 516080 filed on July 18, 2019, describes a technique that allows overcoming these technical limitations so that skin color in an image can be accurately estimated regardless of the lighting conditions.

[0005] Specifically, patent application PCT / US 2020 / 041350 describes a system for generating an automatic estimate of skin color using at least one machine learning model. A user has a mobile computing device equipped with a photographic sensor, which captures one or more images of their face. The mobile computing device transmits these one or more images to a device for determining skin color, which uses one or more machine learning models to determine the skin color based on the images.

[0006] For example, a user can capture multiple images instead of a single image by recording video, thus modifying the lighting conditions by moving the computing device during video capture. Once multiple images under different lighting conditions are provided, a machine learning model can be used to generate multiple determinations of skin color, which can then be averaged or otherwise combined to improve the accuracy of the determination.

[0007] The technology summarized in the aforementioned patent application PCT / US 2020 / 041350 has the following drawback: the user must perform image capture manually, which may be troublesome or even impossible in some cases.

[0008] Therefore, it is hoped that the effort required for users to accurately determine their skin color based on images with different and in principle unknown lighting conditions can be reduced.

[0009] In this regard, a method for determining the skin color of a user's area of ​​interest (e.g., the user's face) is proposed, the method comprising:

[0010] - Import at least one image from the image library of the user's computing device, which includes a representation of the user's region of interest;

[0011] - Process the at least one image using a machine learning model adapted to provide a numerical value for each imported image representing the skin color of the region of interest present in the at least one imported image;

[0012] - Evaluate the skin color of the region of interest based on one or more values ​​from each imported image.

[0013] For example, a user's computing device can be a smartphone, tablet computer, personal computer, smartwatch, or any other computing device capable of having an image library.

[0014] Conventionally, "image library" means a catalog provided on a user's computing device that provides access to all or almost all images and photographs stored in the user's computing device's memory or in remote storage accessible from the user's computing device (e.g., "cloud" type storage).

[0015] The at least one image is preferably a photograph taken by the camera sensor of the user computing device, and optionally a photograph taken by another device and imported into the user computing device.

[0016] It has been noted that, surprisingly, when users provide multiple images from an image library, the machine learning model pre-trained for the techniques summarized in the aforementioned patent application PCT / US 2020 / 041350 achieves even better results, namely, more accurate and more realistic results.

[0017] Specifically, it turns out that, although previously trained to process images whose lighting conditions were modified during video capture due to the movement of the mobile computing device, the machine learning model performed even better when processing images with more diverse and variable lighting conditions (i.e., images that may exist in the user's image library).

[0018] While the invention is particularly advantageous and beneficial for determining facial skin color, it can also be applied to determining the skin color of another area of ​​interest (e.g., another part of the body), especially for applying at least one cosmetic product to the area of ​​interest based on the assessment of the skin color thereon.

[0019] According to one implementation, the at least one image includes metadata that timestamps the creation of the corresponding image (i.e., the date and time the photo was captured); and the evaluation of skin color is performed based on a weighted average of the values ​​of each image according to the current date and the corresponding timestamp metadata.

[0020] Weighting can correspond to assigning a confidence index to an image based on the date the photo was taken, so that, for example, when assessing skin color, older photos are given less weight or more weight to more recent photos.

[0021] According to one implementation, the numerical values ​​of each image are weighted based on the current date and corresponding timestamp metadata to take into account the tonal variations of skin color throughout the year.

[0022] Specifically, skin tone variations (or in other words, tanning) can be significant in some individuals and are naturally (i.e., ignoring artificial tanning methods) closely related to the seasons of the year. Therefore, a mode of implementation could be provided, for example, to give less weight to photographs taken in the opposite season to the current date and more weight to photographs taken in the same season as the current date.

[0023] Advantageously, weighting is performed taking into account the tonal variations of skin color throughout the year, thereby enabling the assessment of skin color to provide a prediction of skin color, such as skin color after the current date.

[0024] For example, this implementation could be provided in a way that gives less weight to photos taken in seasons preceding the current date and more weight to photos taken in seasons following the current date.

[0025] According to one embodiment, the at least one image includes metadata that timestamps the creation of the corresponding image and / or metadata that geolocates the creation of the corresponding image (i.e., GPS coordinates of the location where the photo was captured); and the processing includes preprocessing that includes color temperature correction for each image based on the corresponding metadata.

[0026] Specifically, timestamp metadata and geolocation metadata allow us to infer lighting conditions and the corresponding color temperature to be compensated for. For example, timestamp metadata can indicate whether a photo was taken during the day or at night, while geolocation metadata can further indicate whether the photo was taken indoors or outdoors.

[0027] Advantageously, color temperature correction is further performed on each image based on weather archive data corresponding to the corresponding metadata, thereby allowing estimation of the color temperature of the lighting conditions at the time the corresponding image was created.

[0028] Weather archive data, in particular, can definitively determine the length of day and night, the times of sunrise and sunset, and sunlight conditions for a given date and location. This allows for more accurate estimations of lighting conditions and more refined color temperature corrections.

[0029] According to one embodiment (in which the user's area of ​​interest is the user's face), importing further includes obtaining a reference image representing a reference face of the user, and the processing includes preprocessing, which includes selecting a reference face from at least one image imported from an image library of the user's computing device via face recognition.

[0030] This allows for the avoidance of compromising skin color assessment due to taking into account the different faces of different people during processing, such as in the case of incorrectly imported images or images containing multiple faces.

[0031] For example, obtaining the reference image includes capturing a photograph of a specific area of ​​interest for the user, or having the user identify an image representing that specific area of ​​interest from images in the image library.

[0032] According to one implementation, the machine learning model is a pre-trained convolutional neural network.

[0033] According to one embodiment, the method further includes recommending at least one cosmetic product based on the assessment of the skin color of the area of ​​interest.

[0034] A system for determining the skin color of a user's region of interest is also provided. The system includes a communication device adapted to communicate with a user computing device and configured to import at least one image containing a representation of the user's region of interest from an image library of the user computing device. The system includes a processing device configured to: perform processing on the at least one image using a machine learning model adapted to provide a numerical value representing the skin color of the region of interest present in each imported image; and further perform an evaluation of the skin color of the region of interest based on the one or more numerical values ​​of each imported image.

[0035] A system is also provided, comprising: a communication device adapted to communicate with a user computing device and configured to implement the import step of the method as defined above; and a processing device configured to implement the execution processing and evaluation steps of the method as defined above.

[0036] According to one embodiment, the system further includes a user computing device configured to transmit the at least one image from the image library to a communication device.

[0037] According to one embodiment, the communication device is adapted to communicate with a user's computing device via a telecommunications network (such as the Internet).

[0038] A computer program is also provided, including instructions that, when executed by a computer, cause the computer to perform the methods defined above.

[0039] A computer-readable medium is also provided, including instructions that, when executed by a computer, cause the computer to perform the methods as defined above.

[0040] Other advantages and features of the invention will become apparent from a close reading of the fully non-limiting description of the embodiments and implementations, as well as the accompanying drawings, in which:

[0041] Figure 1 , Figure 2 and Figure 3 The embodiments and implementation modes of the present invention are illustrated.

[0042] Figure 1 An example of a system SYS is shown, which includes a processing unit PU configured to perform processing TS on the at least one image IMi, IMj, IMk using a machine learning model AI_TS and evaluate the skin color of a user-focused region present in one or more images IMi, IMj, IMk using EVAL.

[0043] For convenience, and as a result of a non-limiting arbitrary choice, the description is for the case where the set of "at least one image" consists of multiple images. The implementation is similar in all respects to the case where the set of "at least one image" consists of a single image.

[0044] For the same reason, a description is also provided for the non-restrictive case where the user's focus area is the user's face.

[0045] The machine learning model is configured and trained to provide a numerical value ClrEstm_k representing the skin color of the face present in each image IMk for each image IMk.

[0046] The apparatus for performing the EVAL evaluation is configured to assess facial skin color based on the numerical value ClrEstm_k of each image. For example, the evaluation may include calculations that combine the values ​​ClrEstm_k, such as calculating the mean or median, whether or not weighted.

[0047] The communication device COM is adapted to communicate with a user computing device APP and is configured to receive images IMi, IMj, and IMk from the image library LIB of the user computing device APP in IN_DAT.

[0048] Specifically, the user computing device APP can belong to the system SYS, because the user computing device can be specifically configured to cooperate with the processing unit PU, for example, via software, via the execution of an application or website, so as to transmit these images during the process of the processing unit PU importing images IMi, IMj, IMk into IN_DAT.

[0049] The user computing device (APP) and the communication device (COM) can communicate using any suitable communication technology, such as wireless communication technologies like Wi-Fi, Wi-MAX, Bluetooth, 2G, 3G, 4G, 5G, and LTE, or wired communication technologies like Ethernet, FireWire, and USB. Specifically, the user computing device and the device used to determine skin color can communicate, at least partially, via the Internet.

[0050] The user's computing device app can be, for example, a smartphone, tablet, personal computer, smartwatch, or any other computing device that can have an image library (LIB).

[0051] Typically, the Image Library (LIB) is a directory provided within the user's computing device app's interface that provides access to all or almost all images and photos IM1-IM4, IMk stored in the user's computing device app's internal non-volatile memory INT_NVM. The images and photos IM1-IM4, IMk in the LIB can also, or alternatively, be stored in remote storage on a server CLD accessible from the user's computing device app (e.g., using the well-known expression, stored in the "cloud").

[0052] The at least one image IM1-IM4, IMk is preferably a facial photograph taken by the camera sensor CAM of the user's computing device APP, or a facial photograph taken by another device and imported into the image library LIB of the user's computing device APP.

[0053] Images IM1-IM4 in the image library IMk have digital image data formats, such as JPEG (JPEG stands for Joint Picture Experts Group), PNG (PNG stands for Portable Web Graphics), GIFF (GIFF stands for Graphics Interchange Format), TIFF (TIFF stands for Tag Image File Format), or any other image format.

[0054] Advantageously, each image IM1-IM4, IMk in the image library IMk further includes metadata MTD1-MTD4, MTDk that provides various information about the image.

[0055] Specifically, at least some of the images IM1-IM4 and IMk in the library LIB include timestamp metadata that provides information about the creation date and time of the respective image, and / or geolocation metadata that provides information about the creation location of the respective image.

[0056] As shown in the reference below Figure 2 The timestamp metadata and / or geolocation metadata that we see can enable the processing performed by the machine learning model AI_TS and the evaluation of skin color EVAL to be refined and improved.

[0057] The skin color obtained through the machine learning model AI_TS is so realistic that it can then be used to recommend one or more cosmetic products that are precisely related to each user's skin color, especially foundation, pressed powder, variations of foundation and pressed powder, and cosmetic products that match or suit the user's skin color.

[0058] Please refer to the following: Figure 2 .

[0059] Figure 2 An example of a method for determining facial skin color is shown, which specifically uses a reference... Figure 1 The system described is implemented by the processing unit PU.

[0060] The method includes, for example, referencing Figure 1 The method described imports at least one image IMk from the image library LIB of the user's computing device into IN_DAT.

[0061] The processing TS implemented in this example for the at least one image IMk includes preprocessing P_TS and processing using a machine learning model AI_TS, which is configured to provide a numerical value ClrEstm_k representing the skin color of a face present in the at least one image IMk.

[0062] Preprocessing P_TS involves converting the image IMk into a format suitable for the machine learning model AI_TS. Specifically, preprocessing P_TS is suitable for separating one or more parts of the image containing the face via the conventional face detection mechanism FceDet, which is known in itself, and for centering and scaling the parts of the image containing the face using Cntr+Scl, in order to provide standardized data to the machine learning model.

[0063] In short, centering (Cntr) can include cropping the image to retain only the portion containing the face, while scaling (Scl) includes enlarging or reducing the cropped image and downsampling or oversampling the pixels of the cropped image to provide a cropped image with a set size and set resolution.

[0064] Advantageously, the preprocessing further includes selecting cropped images containing the same face via the conventional face recognition mechanism FceReco, which is known per se.

[0065] The face recognition function FceReco identifies and detects the target face, specifically the reference face provided in the reference image IMref. Advantageously, the reference face is isolated within the reference image IMref, meaning it is the only face present in the reference image IMref.

[0066] The reference image IMref can be imported, for example, from the image library LIB of the user's computing device and recognized by the user as the reference image IMref. Alternatively, the reference image IMref can be taken by the user through the photographic sensor (CAM) of the user's computing device, and it is advantageous that it is a self-portrait photograph (often referred to as a "selfie") PhtSlf, so as to include the individual user face as the reference face.

[0067] As referenced above Figure 1 The image IMk in the image library advantageously includes metadata MTDk. In this case, preprocessing P_TS can advantageously include color temperature correction TempCorr on the imported image, which is established based on the corresponding metadata MTDk.

[0068] Color temperature is a quantity well known to those skilled in the art and is usually measured in Kelvin. It characterizes a light source by comparing it with the theoretical principle of blackbody thermal radiation.

[0069] Advantageously, color temperature correction for each image is further performed based on external data such as weather archive data and geographic data EXT_MTD, thereby allowing the color temperature of the lighting conditions at the time of creating the corresponding image to be estimated in accordance with the corresponding metadata MTDk.

[0070] Specifically, timestamp metadata (MTDk ts) and geolocation metadata (MTDk geoloc) allow for the estimation of lighting conditions and the corresponding color temperature to be compensated for. For example, timestamp metadata can indicate whether a photo was taken during the day or at night, while geolocation metadata can further indicate whether the photo was taken indoors or outdoors.

[0071] Weather archive data, in particular, can definitively determine the length of day and night, the times of sunrise and sunset, and sunlight conditions for a given date and location. This allows for more accurate estimations of lighting conditions and more refined color temperature corrections.

[0072] The color temperature correction TempCorr performed by the preprocessing unit P_TS can, for example, use a list of conditions that the metadata MTDk may satisfy, while the external data EXT_MTD can optionally be indexed with the metadata MTDk.

[0073] Lookup tables may allow selection of specific color temperature corrections based on which conditions are met or not met.

[0074] For example, among the simple possible conditions, one could mention "dawn," "daytime," "night," "dusk," "indoors," "outdoors," "sunny day," "cloudy day," "rainy day," "snowy day," etc. In this regard, one can imagine and conceive of other conditions as well as more complex ones.

[0075] For example, if the metadata MTDk reflects "night" and "indoor" conditions, then color temperature correction may be performed to correct for "indoor lighting" or "incandescent" type lighting. If the metadata MTDk reflects "dusk," "outdoor," and "sunlight" conditions, then color temperature correction may be performed to correct for sunset type lighting.

[0076] The mechanism of color temperature correction TempCorr is similar to debayering or demosaicing, a technique typically used to rebalance the RGB channels of a RAW image captured by a photographic sensor individually before storing the image.

[0077] Therefore, the color temperature correction TempCorr mechanism advantageously uses a deBayer matrix or demosaicing matrix, the points of which are customized based on conditions evaluated in the imported image IMk via metadata MTDk. These adjustments are then integrated into a static image file, which is validated by assuming that the color of an element in the image is always known (e.g., eyes are white, or to a lesser extent, teeth are white).

[0078] The image IMk, which is then centered by FceDet (Cntr), scaled by Scl, optionally selected by FceReco, and optionally corrected by TempCorr, is provided to the machine learning model AI_TS.

[0079] The machine learning model AI_TS calculates the numerical value ClrEstm_k representing the skin color of the face present in each image IMk.

[0080] It can be done, for example, by the method described in patent application PCT / US 2020 / 041350 or as referenced below. Figure 3 The summary describes the implementation of the machine learning model AI_TS using a pre-trained convolutional neural network. The machine learning model AI_TS can also be a feedforward neural network or a recurrent neural network. Any suitable training technique can be used, particularly gradient descent techniques such as stochastic gradient descent, batch gradient descent, and mini-batch gradient descent.

[0081] In addition, the machine learning model AI_TS may be able to select a suitable image ImSel by detecting anomalous image capture conditions in the image (such as particularly unsuitable image capture angles).

[0082] The selection of such eligible images can be performed in two stages. First, the machine learning model allows for explicit testing of the met conditions (e.g., face facing the camera sensor, only a single face in the image, face actually present in the image). After this first filtering stage, a second selection is performed using a quality score wght1, which is learned, for example, in a weakly supervised manner during the training of the model AI_TS. The quality score wght1 is purely statistical and is optimized, particularly during training, to improve the accuracy of the model AI_TS.

[0083] Therefore, the machine learning model AI_TS can assign a first weight wght1 to each result, especially allowing the discarding of anomalous results.

[0084] When the machine learning model AI_TS has provided a value ClrEstm_k for each image IMk, the facial skin color evaluation EVAL is performed based on the value ClrEstm_k and the corresponding first weight wght1.

[0085] For example, evaluating EVAL may include calculations that combine values ​​ClrEstm_k, such as using a first weight wght1 to calculate the mean or median.

[0086] Furthermore, the valuation can be advantageously weighted by a second weight wght2 obtained based on the metadata MTDk (especially timestamp metadata) of the corresponding image, in order to take into account the tonal changes of skin color throughout the year, i.e., skin tanning.

[0087] Specifically, it is expected that photographs taken in the same season as the current date (i.e., the date on which the facial skin color was determined using the disclosed technique) can be selected.

[0088] For example, a lower value of the second weight wght2 can be used to reduce the weight of an image whose season is opposite to the current date, while a higher value of the second weight wght2 can be used to increase the weight of an image whose season is the same as the current date.

[0089] Advantageously, a second weight wght2 can be determined, thereby enabling the skin color assessment EVAL to provide a prediction of skin color, such as skin color after the current date.

[0090] For example, this can be achieved by giving less weight to images whose season is earlier than the current date's season, and more weight to images whose season is later than the current date's season.

[0091] The relationship between skin color and season was measured using a spectral colorimeter (or "spectrometer"), revealing the average skin color variation that occurs in summer relative to winter. These measurements established macroscopic variations within the population and were integrated into a model for evaluating EVAL in order to determine the second weight wght2.

[0092] Figure 3 A non-limiting example of an implementation that trains a machine learning model AI_TS to a TRN to provide a numerical value ClrEstm_k representing the skin color of at least one face present in an image IMk is shown, as referenced above. Figure 1 and Figure 2 As stated above.

[0093] Training can be implemented by a processing unit PU that includes an additional training unit TRN, which is specifically configured to control the unit IN_DAT for receiving data, the preprocessing unit P_TS, the machine learning model AI_TS, and the evaluation unit EVAL.

[0094] A set of training images IN_DAT associated with real skin color information GRND_TRTH (referred to as "real data") is collected to implement the training TRN on the machine learning model AI_TS.

[0095] The real data on skin color, GRND_TRTH, was used as empirical evidence or information to label images provided by the training subjects (i.e., volunteers, referred to as group members).

[0096] For example, real-world data GRND_TRTH can be collected by users using industry-standard techniques for determining skin color, such as comparison with colorimetric charts or evaluation by a spectrophotometer specifically designed for measuring skin color.

[0097] In comparison with a colorimetric chart, a portion of the image may contain a representation of a known reference colorimetric chart. Correcting the image colors to restore the representation of the reference colorimetric chart to its original colors can allow for the determination of true data GRND_TRTH regarding the skin colors present in the color-corrected image.

[0098] For example, color correction based on real data GRND_TRTH can be implemented in the preprocessing P_TS according to the commands of the training device TRN.

[0099] A spectrophotometer can be used on the skin of a group member at least once, and the color measured by the spectrophotometer (regardless of any color code transformation) can be used directly as the real data GRND_TRTH.

[0100] The set of training images IN_DAT was obtained by the group members through reference above. Figure 1 The device is provided by a user computing device of the same type as the APP.

[0101] For example, the training image IN_DAT is imported via a photosensitive sensor CAM on a user computing device used by the group members, which is used to capture one or more training images including the faces of the group members. The training images can be extracted from, for example, a personal photograph PhtSlf taken in selfie mode, or from, for example, at least one video recording Vid360 taken in selfie mode, in which lighting conditions may have been modified due to perspective shifts (e.g., by rotating the device around the faces of the group members).

[0102] Multiple training images can then be generated by extracting personal images from the Vid360 video.

[0103] Capturing video (from which multiple training images can be extracted) may be advantageous, at least because it would greatly increase the efficiency of generating large amounts of training data under various lighting conditions.

[0104] If necessary, a person skilled in the art may refer to the training description of the machine learning model described in patent application PCT / US 2020 / 041350.

[0105] It should be noted that during the training process, the preprocessing P_TS only includes the components mentioned above. Figure 2 The described face detection method is FceDet, and centering and scaling are performed using Cntr+Scl. Specifically, it can be assumed that group members only provide their own facial images; in this case, face recognition (FceReco) is not required. Furthermore, given that the training images are extracted from photos or videos, the metadata does not provide a specific context that allows parameterization of the color temperature correction (TempCorr).

[0106] Optionally, training images IN_DAT can be additionally or alternatively extracted from the image library (IMk∈LIB) of the user computing devices of the group members. In this case, the execution can be specified as described in the reference. Figure 2 The steps described are: face recognition FceReco and color temperature correction TempCorr.

[0107] Therefore, it has been noted, surprisingly, that training using only photographic and video recordings (such as those described above or actually as described in patent application PCT / US 2020 / 041350) in reference Figure 2 The described implementation (i.e., the machine learning model AI_TS uses images from the image library LIB for inference) provides very satisfactory results.

[0108] During training, the machine learning model AI_TS performs parameterizable computations on the training image IN_DA and provides a numerical value ClrEstm representing the skin color of the faces present in the training image.

[0109] The assessment of skin color, EVAL, can be performed via statistical calculations AVRG that combine various values ​​(ClrEstm), as shown in the reference. Figure 2 As mentioned above.

[0110] Then, the skin color (or actually the numerical value ClrEstm) evaluated by EVAL is compared with the real data GRND_TRTH to reparameterize the computation performed by the machine learning model AI_TS, so as to get as close to the real data as possible.

[0111] In other words, the real data about skin color, GRND_TRTH, is used as labeled data to instruct the machine learning model AI_TS on the expected results of processing the training image IN_DAT.

[0112] It should be noted that it is advantageous to train the machine learning model AI_TS using data from multiple group members. In this regard, the training image IN_DAT and the real data GRND_TRTH associated with each group member can be stored in the memory of the processing unit PU and the training unit TRN. The computation of the machine learning model AI_TS is parameterized for all data from all group members in order to be as universal as possible.

[0113] In short, to determine facial skin color, the processing device PU uses the machine learning model AI_TS, as described above, to determine the facial skin color present in at least one image imported from the image library of the user's computing device APP. This skin color can then be used to recommend one or more cosmetic products that match or suit the determined skin color.

Claims

1. A method for determining the skin color of a user's area of ​​interest, the method comprising: - Import at least one image (IMk) including a representation of the user's region of interest from the image library (LIB) of the user's computing device (APP); - Obtain at least one numerical value (ClrEstm_k), each of the at least one numerical value (ClrEstm_k) being associated with a corresponding image (IMk) of the at least one image (IMk); - The at least one numerical value (ClrEstm_k) is obtained by performing processing (TS) on the at least one image (IMk) using a machine learning model (AI_TS), the machine learning model being adapted to provide a numerical value (ClrEstm_k) representing the skin color of the region of interest present in the imported image (IMk) for each imported image (IMk). - Perform an evaluation (EVAL) on the skin color of the region of interest based on at least one of the values ​​(ClrEstm_k); in: - The at least one image (IMk) includes timestamp metadata that timestamps the creation of the corresponding image and geolocation metadata that geolocates the creation of the corresponding image; - The processing (TS) includes preprocessing (P_TS), which includes color temperature correction (TempCorr) of each image (IMk) based on the corresponding timestamp metadata and geolocation metadata. The color temperature correction includes determining the location type associated with the creation of the image from the geolocation metadata. The location type is either indoor space or outdoor space.

2. The method according to claim 1, wherein, The area of ​​user attention is the user's face.

3. The method according to claim 1, wherein: - The evaluation of the skin color (EVAL) is performed based on a weighted average (wght2) of the numerical value (ClrEstm_k) of each image according to the current date and corresponding timestamp metadata.

4. The method according to claim 3, wherein, The numerical value (ClrEstm_k) of each image is weighted (wght2) based on the current date and corresponding timestamp metadata to take into account the tonal variations of the skin color throughout the year.

5. The method according to claim 4, wherein, The weighting (wght2) is performed taking into account the tonal variations of the skin color throughout the year, so that the evaluation (EVAL) of the skin color on dates after the current date provides a prediction of the skin color.

6. The method according to claim 1, wherein, Based on the weather archive data (ArchMeteo) corresponding to the corresponding timestamp metadata and geolocation metadata, the color temperature correction (TempCorr) is further performed on each image, thereby allowing estimation of the color temperature of the lighting conditions at the time the corresponding image was created.

7. The method according to claim 2, wherein: - The import further includes obtaining a reference image (IMref) representing the user's reference face; - The process includes preprocessing (P_TS), which includes selecting a reference face from the at least one image (IMk) imported from the image library of the user computing device via face recognition (FceReco).

8. The method according to claim 7, wherein, Obtaining the reference image (IMref) includes capturing a photograph (PhtSlf) of a single region of interest for the user, or having the user identify an image representing the single region of interest from images (IMk) in the image library.

9. The method according to claim 1, wherein, The machine learning model (AI_TS) is a pre-trained (TRN) convolutional neural network.

10. The method of claim 1, further comprising recommending at least one cosmetic product based on the assessment (EVAL) of skin color in the area of ​​interest.

11. A system for determining the skin color of a user's area of ​​interest, the system comprising: - A communication device (COM) adapted to communicate with a user computing device (APP) and configured to import at least one image (IMk) including a representation of the user's region of interest from the image library (LIB) of the user computing device (APP), wherein the at least one image (IMk) includes timestamp metadata that timestamps the creation of the corresponding image and / or geolocation metadata that geolocates the creation of the corresponding image; - A processing device (PU) configured to obtain at least one numerical value (ClrEstm_k), each of the at least one numerical value (ClrEstm_k) being associated with a corresponding image (IMk) of the at least one image (IMk). The processing device (PU) is configured to perform processing (TS) on the at least one image (IMk) using a machine learning model (AI_TS) to obtain the at least one numerical value (ClrEstm_k). The machine learning model is adapted to provide a numerical value (ClrEstm_k) representing the skin color of the region of interest present in the at least one imported image (IMk) for each imported image (IMk). The processing (TS) includes preprocessing (P_TS), which includes color temperature correction (TempCorr) of each image (IMk) based on corresponding timestamp metadata and geolocation metadata. The color temperature correction includes determining the location type associated with the creation of the image from the geolocation metadata. The location type is either indoor space or outdoor space. - The processing device (PU) is also configured to perform an evaluation (EVAL) of the skin color of the region of interest based on the at least one numerical value (ClrEstm_k) of each imported image.

12. The system according to claim 11, wherein, The communication device (COM) is further configured to implement the introductory step of the method according to any one of claims 2 to 10, and the processing device (PU) is further configured to implement the execution processing (TS) and evaluation (EVAL) steps of the method according to any one of claims 2 to 10.

13. The system of claim 11, further comprising the user computing device (APP), the user computing device being configured to transmit the at least one image (IMk) of the image library (LIB) to the communication device (COM).

14. The system according to claim 11, wherein, The communication device (COM) is adapted to communicate with the user computing device (APP) via a telecommunications network (RES).

15. A computer-readable medium comprising instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1 to 10.

Images