A method for semantic facial image manipulation and aesthetic treatment recommendation based on perceptual features

Abstract

One aspect is a method of generating an image of a human face that is expected to have at least one altered perceptual feature. Such a method includes encoding an image of a human face into a vector to generate an input image vector. A scaled directional vector is applied to the input image vector to generate a modified input image vector. The modified input image vector is transformed into a deformed image of a real human face. The scaled directional vector is determined using an artificial neural network trained to associate evaluations of human faces and perceptual features to determine unit directional vectors. The unit directional vector is based on at least one selected perceptual feature. The amplitude for scaling the unit directional vector is determined by the limits of facial feature change obtainable by known treatments.

Description

2. Background of the Invention

[0001] The present disclosure relates to the field of processing images of human anatomical features, such as faces, to predict expected facial appearances as a result of potential treatments. Such treatments may include cosmetic, surgical, or medical treatments, such as injections of fillers or other agents, although the present disclosure is not limited to medical treatments. More specifically, the present disclosure relates to methods for creating predicted images based on machine learning processing of actual changes in appearance resulting from facial treatments on real people. The present disclosure further relates to methods for recommending suitable treatments to obtain improvements in one or more perceptual features associated with a human face.

[0002] Cosmetic treatments, such as medical cosmetic treatments, as well as surgical treatments and injections (e.g., but not limited to, Botox and fillers), are performed to modify the appearance of human features, such as the face. Although physicians, particularly plastic surgeons, have detailed knowledge of how such procedures are performed and have an informed understanding of how such procedures change the appearance of human features, it is difficult to convey to the individual what may be reasonably expected as a result of such procedures. Furthermore, while physicians may have the ability to select the amount of change in a feature relative to the strength of the treatment and the products used, they may still be unable to convey to the individual how variations in the amount of change in the feature will affect the overall appearance when the treatment is completed.

[0003] The appearance of a human face can create quantifiable perceptions in a human observer. US Patent No. 6,239,993 issued to Velthuis et al. discloses a system that receives an image of a human face as input and uses a set of image processing and deep learning algorithms to analyze the image to determine landmarks on the face that are characteristic of predetermined anatomical parts of the face. The system compares the landmarks to a model generated based on faces scored for a plurality of perceptual features through scientifically validated surveys of people. The system uses the model to determine a score for each of the plurality of perceptual features of the face based on the comparison. The system uses the model to determine a first impression of the face based collectively on the scores of all the perceptual features determined by the model for the face. The system provides an output that includes the first impression and the scores of the perceptual features determined by the model for the face.

[0004] In addition to the system disclosed in the above-mentioned US Patent No. 6,239,663, other methods known in the art provide methods for modifying an input image to display the expected appearance of a human face after one or more such treatments. What is needed are tools that allow for suggesting suitable treatments to improve an observer's assessment of one or more perceptual features of a human face. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] U.S. Patent No. 11,151,362 Summary of the Invention

[0006] One aspect of the present disclosure is a method for generating an image of a human face that is expected to have at least one altered perceptual feature. The method according to this aspect of the present disclosure includes encoding an image of a human face into a vector to generate an input image vector. A scaled directional vector is applied to the input image vector to generate a modified input image vector. The modified input image vector is transformed into a transformed image of a real human face. The scaled directional vector is then applied to the input image vector to generate a modified input image vector. The selected directional vector is determined using an artificial neural network trained to associate evaluations of human faces and perceptual features to determine a unit directional vector, the unit directional vector being based on at least one selected perceptual feature, and an amplitude for scaling the unit directional vector is determined by the limits of change in the facial feature obtainable by any available treatment or facial adjustment.

[0007] In some embodiments, the unit direction vector is determined by selecting a highest rated human face evaluation image and a lowest rated human face evaluation image from among a plurality of human face evaluation images, said evaluation being performed by an artificial neural network for at least one selected perceptual feature, and determining a direction of the unit direction vector from a vector representation of the selected highest rated image and a vector representation of the selected lowest rated image.

[0008] In some embodiments, the evaluation images of human faces include randomly generated artificial human face images, where the randomly generated artificial human face images are generated by a generative adversarial network.

[0009] Some embodiments further include using the input image and the deformed image to automatically generate at least one recommended treatment to make the human face from which the input image was created most similar to the deformed image.

[0010] In some embodiments, automatically generating at least one recommended treatment includes inputting the input image and the deformed image into an artificial neural network trained on change detection to detect particular changed regions in the input image, comparing the detected changed regions to facial region masks known to be affected by particular treatments, and adding the particular treatment to a list of suggested treatments if an intersection of at least one of the detected changed regions with at least one of the facial region masks is greater than a predefined threshold.

[0011] In some embodiments, the change regions are detected by computing a difference image between the actual human face image and the deformed image.

[0012] In some embodiments, the change regions are detected by inputting images of actual human faces and the deformed images into an artificial neural network trained in change detection.

[0013] In some embodiments, automatically generating the at least one recommended treatment includes inputting the input image and the transformed image into an artificial neural network that has been trained by inputting pre-treatment and post-treatment images of actual human faces that have received the at least one treatment.

[0014] In some embodiments, automatically generating at least one recommended treatment includes comparing a directional vector of the input image with reference to the deformed image to a directional vector of each of a plurality of treatments, and selecting one of the plurality of treatments that most closely matches the directional vector of the input image with reference to the deformed image.

[0015] In some embodiments, the encoded image is an actual human face.

[0016] According to another aspect of the present disclosure, a computer program is stored on a non-transitory computer readable medium, the program including logic operable to cause a programmable computer to perform acts including: encoding an image of a human face into a vector; An input image vector is generated. The scaled direction vector is applied to the input image vector to generate a modified input image vector. The modified input image vector is transformed into a deformed image of a real human face. The scaled direction vector is determined by training a first artificial neural network that determines a change in appearance of the human face image corresponding to a change in the perceptual feature to determine a direction vector based on at least one selected perceptual feature, and an amplitude for scaling the direction vector is determined by a limit of the change in the facial feature obtainable by a known treatment.

[0017] In some embodiments, the unit direction vector is determined by selecting a highest rated human face evaluation image and a lowest rated human face evaluation image from among a plurality of human face evaluation images, said evaluation being performed by an artificial neural network for at least one selected perceptual feature, and determining a direction of the unit direction vector from a vector representation of the selected highest rated image and a vector representation of the selected lowest rated image.

[0018] In some embodiments, the evaluation images of human faces include randomly generated artificial human face images, where the randomly generated artificial human face images are generated by a generative adversarial network.

[0019] Some embodiments further include using the input image and the deformed image to automatically generate at least one recommended treatment to make the human face from which the input image was created most similar to the deformed image.

[0020] In some embodiments, automatically generating at least one recommended treatment includes inputting the input image and the deformed image into an artificial neural network trained on change detection to detect particular changed regions in the input image, comparing the detected changed regions to facial region masks known to be affected by particular treatments, and adding the particular treatment to a list of suggested treatments if an intersection of at least one of the detected changed regions with at least one of the facial region masks is greater than a predefined threshold.

[0021] In some embodiments, the change regions are detected by computing a difference image between the actual human face image and the deformed image.

[0022] In some embodiments, the change regions are detected by inputting images of actual human faces and the deformed images into an artificial neural network trained in change detection.

[0023] In some embodiments, automatically generating the at least one recommended treatment includes inputting the input image and the transformed image into an artificial neural network that has been trained by inputting pre-treatment and post-treatment images of actual human faces that have received the at least one treatment.

[0024] In some embodiments, automatically generating at least one recommended treatment includes comparing a directional vector of the input image with reference to the deformed image to a directional vector of each of a plurality of treatments, and selecting one of the plurality of treatments that most closely matches the directional vector of the input image with reference to the deformed image.

[0025] In some embodiments, the encoded image is an actual human face.

[0026] Other aspects and possible advantages will become apparent from the following description and claims. . [Brief description of the drawings]

[0027] [Figure 1] 1 shows a flowchart of a process for training a neural network to determine directional vectors for transforming a human face image to improve the expected assessment of one or more selected perceptual features. [Diagram 2] 1 is a flowchart of a process for deforming an input image of a real human face in a manner that results in an improvement of one or more selected perceptual features. [Diagram 3] 1 is a flow chart of an example of an embodiment of a treatment recommendation method. [Figure 4] 1 is a flow chart of an example of an embodiment of a treatment recommendation method. [Diagram 5] FIG. 1 is a schematic diagram of an example of a computer system that may be used in some embodiments. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0028] The present disclosure describes a procedure for generating realistically modified human face images that are expected to exhibit different (higher or lower) scores in one or more selected perceptual features, and a recommender that suggests treatments, e.g., surgical, cosmetic, and other medical treatments, that are expected to result in changes to a human face, e.g., a real human face, that will result in changes, e.g., improvements, to one or more perceptual features when the treated human is observed by other people. In this disclosure, the term "image" is not limited to still images. Moving images, e.g., videos, are fully within the scope of the present disclosure.

[0029] The procedure is based on the use of neural networks such as Convolutional Neural Networks (CNNs), a training set of face images corresponding to perceptual feature labels, and Generative Adversarial Networks (GANs).

[0030] An example of a method embodiment uses a GAN trained on face images. The GAN is pre-trained to generate random images of realistic looking human faces. The images generated by the GAN are used in the disclosed method to identify and evaluate one or more perceptual features and to determine latent space directions and vector amplitudes for transforming a face image, e.g., an image of a natural person, that is used as an input to improve one or more perceptual features. The method then determines one or more treatments that may result in a change to the human face that results in an improvement in one or more perceptual features. As used in this disclosure, the term "treatment" is intended to include any drug action, structural change, or manipulation, including but not limited to surgical, cosmetic, or other medical treatments, manipulations, color changes, Fitzpatrick skin type changes, etc., that are expected to result in a change to the appearance of the human face, whether permanent or temporary. Thus, the specific examples of treatments provided herein are intended only to provide examples of possible embodiments and are not intended to limit the scope of the disclosure in any way.

[0031] Methods or procedures according to the present disclosure may be implemented on a computer or computer system, as further described below. A non-transitory computer-readable medium may store thereon instructions or logic operable to program a programmable computer or computer system to perform a method or procedure according to the present disclosure.

[0032] An example embodiment of a method or procedure according to the present disclosure may begin with training a neural network, such as a convolutional neural network (CNN), to automatically evaluate one or more perceptual features of an image, such as an image of a human face. In this example embodiment, the CNN may be trained using survey results from human observers who are shown images of human faces and indicate their subjective evaluation of the one or more perceptual features. An example of such a form is described in U.S. Patent No. 11,151,362 issued to Velthuis et al., which is incorporated herein by reference. The present embodiment describes training of a CNN, although a pre-trained CNN may be used, provided that the CNN is pre-trained using procedures as disclosed herein to include an automated assessment of the specific perceptual features to be improved. The trained CNN may be based on a validated database of a compilation of a large number of people's faces that have been assessed or scored by other people (i.e., a broad spread of people with respect to their demographics) from diverse backgrounds, socio-economic statuses, age groups, etc. The faces are scored based on a number of perceptual features. Perceptual features include, but are not limited to, personality attributes (e.g., dominant, attractive, kind, etc.) and socio-economic factors (e.g., rich, fashionable, etc.) that are associated with a person's face when forming a first impression of that person's face. Non-limiting examples of perceptual features include attractiveness, competence, dominance, intelligence, warmth, trustworthiness, etc.

[0033] 1, the trained GAN, as described above, may be used to randomly generate a set of face images, at block 10. The set of face images may include on the order of 100,000 to 1,000,000 individual images of different human faces generated by the GAN, although the exact number of images generated is not intended to limit the scope of this disclosure.

[0034] At block 12, each image in the set of random face images generated by the GAN may be evaluated by the trained CNN for one or more perceptual features. At block 14, a predetermined number of images may be selected from among the evaluated images as representative of the highest and lowest values ​​of the one or more selected perceptual features. The N×D vectors of high and low rated images may be used as inputs to train a binary classifier. At block 14, a decision boundary is determined, with the N×D vectors of high rated images being placed on one side of the decision boundary and the N×D vectors of low rated images being placed on the other side of the decision boundary. A vector perpendicular to the decision boundary determined at block 16 represents a direction to apply to the N×D dimensional encoded image vector of an image of a real human face (the “input image”) to improve the one or more selected perceptual features.

[0035] Once the orientation of the N×D image vector is determined, the above procedure then determines the amplitude associated with the selected one or more perceptual features. In some embodiments, the amplitude may be determined by iteratively varying the amplitude, vector-adding it to an N×D vector representation of the image vector, and sending the manipulated image vector representation to a trained GAN to generate a manipulated image. The above procedure may be repeated with incremental changes in the amplitude until a final value of the amplitude is reached.

[0036] The amplitude corresponds to the strength of the change in facial appearance. The amplitude determination process is performed to identify the maximum amplitude that results in a facial change that preserves the facial identity, such change being obtained by aesthetic, cosmetic, or other types of treatment. The identified value serves as a natural (social) limit for facial changes. Those skilled in the art will understand that at some amount of change, the deformed facial appearance is outside the range of subjectively acceptable human facial appearances when observed by a real human.

[0037] In some embodiments, the final value of the amplitude may be limited by entering a hard-coded upper limit value of the amplitude into the computer or computer system. In some embodiments, the hard-coded upper limit value may be determined as follows:

[0038] In some embodiments, the change in landmarks may be used: The images are used to assess changes in facial landmarks associated with the areas of the face affected by the treatment, both before and after the treatment is administered. When the difference or ratio between selected landmarks reaches a predefined limit, further landmark modification is halted and the amplitude value at that point is used. If the predefined limit is not reached, the landmark may be modified, but the process returns to determining whether to modify the landmark. The ratio or difference may be understood to relate to anatomical features, such as distances between various specific points on the human face. It is understood by medical practitioners that certain distances or ratios between certain anatomical features lie within empirically determined limits. Such empirically determined limits may be used as limits for the amplitude determination.

[0039] In some embodiments, when the change in color and texture of the affected area reaches a predefined limit, the process is stopped and the amplitude value at that point is used as the final amplitude value. The predefined limits for landmarks and colors may be obtained, for example, using pre- and post-treatment images of real human anatomical features.

[0040] In some embodiments, constraints on amplitude values ​​may be defined manually by a subject matter expert after viewing multiple manipulated face images with various amplitude values.

[0041] In some embodiments, additional constraints on the amplitude values ​​may be defined automatically by a CNN trained in face recognition.

[0042] Once the direction and magnitude for adjusting the input image vector have been determined, the result is a scaled directional vector. The transformed image may be generated by vector-adding the scaled directional vector to a vector representation of the input image. The scaled directional vector may be determined for any one or more selected perceptual features, and applying the scaled directional vector to an input image of a human face produces a transformed image that is expected to improve the one or more selected perceptual features.

[0043] Referring to Figure 2, at block 18, an image of a real human face may be deformed using a GAN trained as described with reference to Figure 1. At block 20, an input image may be encoded into an NxD vector to form an input image vector. At block 22, a scaled direction vector may be applied to the input image vector. At block 24, a deformed image may be generated using the input image vector and the scaled direction vector.

[0044] In block 26, the deformed image may then be used as input by a process that generates one or more treatment suggestions that, when applied to the human face used to generate the input image, will most faithfully resemble the deformed image.

[0045] In some embodiments, and with reference to FIG. 3, change regions in a human face may be detected by unsupervised or supervised change detection methods: in an exemplary unsupervised embodiment of a change detection method, in block 30 a difference image, e.g. a grayscale difference image, may be calculated by subtracting a deformed image from an input image, in block 32 an automatic thresholding is applied to the difference image to detect candidate regions in the image, and in block 34, after deformation operations on the candidate regions such as opening / closing and filtering by size, which are optional operations that may be included in some embodiments, the final regions are identified.

[0046] At block 36, the regions identified as detected change regions are then compared to a facial region mask known to be affected by specific treatments. As shown at block 38, if the percentage of intersections between the change region and the region mask is higher than a predefined threshold, a treatment known to affect the region mask is suggested. The suggested treatment(s) (e.g., in a list for multiple recommendations) may be ordered with respect to intersection percentage before being presented to the user.

[0047] In an exemplary supervised embodiment of the change detection method, referring to Figure 4, an artificial neural network trained for change detection is used to detect specific regions that have changed in a facial image. This is shown in block 40 of Figure 4.

[0048] At block 42, the regions identified within the detected change region are then compared to a facial region mask known to be affected by a particular treatment. If the percentage of intersection between the change region and the region mask is higher than a predefined threshold, the treatment known to affect the region mask is added to a suggested treatment list, as shown at block 44. The suggested treatment(s) (e.g., in a list for multiple recommendations) may be ordered with respect to the intersection percentage before being presented to the user.

[0049] In one embodiment, a neural network, such as a CNN, may be trained on a dataset that includes pre- and post-treatment images of real human faces that have undergone a particular treatment. Once trained, the CNN may function as a classifier between two images of a human face that may determine whether and what treatment has been applied to the human face. The original and deformed face images are input to the trained CNN to suggest treatment(s) to a person considering treatment(s) to obtain a facial appearance that corresponds to an image with a high value(s) of the selected perceptual feature(s). The trained CNN may then output a treatment suggestion such that the original image is closest to the deformed image.

[0050] In one embodiment, the identified perceptual feature direction, i.e., the vector direction that changes one image in a particular way to change the selected perceptual feature, may be compared to previously identified directions for various treatments. The comparison may be performed using the exact identified direction or using extracted components of the identified direction. If the direction of the treatment is found to be similar to the identified (perceived) direction or one of its components, the treatment may be suggested to the user.

[0051] Although the present disclosure refers to increasing the value of at least one perceptual feature, it should be understood that it is within the scope of the present disclosure to use the same methods, and computer programs implementing the methods, to decrease the value of one or more perceptual features.

[0052] FIG. 5 illustrates an exemplary embodiment of a computing system 100 that may be used to perform the operations described with reference to FIGS. 1-4. Any form of storage medium readable by various portions of the computing system 100 may include a computer program having logic operable to cause the computing system 100, or portions thereof, to perform the operations as described with reference to FIGS. 1-4. The computing system 100 may be an individual computer system 101A or an array of distributed computer systems. The individual computer systems 101A may include one or more analysis modules 102 that may be configured to perform various tasks according to some embodiments, such as the tasks described with reference to FIG. 5. To perform these various tasks, the analysis modules 102 may operate independently or in cooperation with one or more processors 104, which may be connected to one or more storage media 106. A display device, such as any known type of graphical user interface, may be connected to the processor 102 to allow input of commands and / or data by a user and to display the results of execution of a set of instructions according to the present disclosure. 104.

[0053] The processor(s) 104 may also be connected to a network interface 108 such that individual computer system 101A can communicate with one or more additional individual computer systems and / or computing systems, such as 101B, 101C, and / or 101D, over a data network 110. (Note that computer systems 101B, 101C, and / or 101D may or may not share the same architecture as computer system 101A; e.g., computer systems 101A and 101B may be located at a well drilling site while communicating with one or more computer systems, such as 101C and / or 101D, located in one or more data centers and / or located in various countries on different continents.)

[0054] A processor may include, but is not limited to, a microprocessor, a microcontroller, a processor module or subsystem, a programmable integrated circuit, a programmable gate array, or other control or computing device.

[0055] Storage medium 106 may be implemented as one or more computer-readable or machine-readable storage media, e.g., non-transitory storage media. The storage medium may include logic operable to cause a computer system to perform the operations described above with reference to Figures 1-4. In the exemplary embodiment of Figure 5, storage medium 106 is shown disposed within individual computer system 101A, but it should be noted that in some embodiments storage medium 106 may be distributed within and / or across multiple internal and / or external enclosures of individual computing system 101A and / or additional computing systems, e.g., 101B, 101C, 101D. The storage medium 106 may include one or more different forms of memory, including, but not limited to, semiconductor memory devices such as dynamic or static random access memory (DRAM or SRAM), erasable and programmable read-only memory (EPROM), electrically erasable and programmable read-only memory (EEPROM) and flash memory, magnetic disks such as fixed disks, floppy disks and removable disks, other magnetic media including tape, optical media such as compact disks (CDs) or digital video disks (DVDs), or other types of storage devices. It should be noted that the computer instructions for causing any individual computer system or computing system to perform the tasks described above may be provided on one computer-readable or machine-readable storage medium, or on multiple computer-readable or machine-readable storage media distributed across multiple component computing systems having one or more nodes. Such computer-readable or machine-readable storage media or media may be considered to be part of an article (or article of manufacture). An article or article of manufacture may refer to a single piece or multiple pieces that are manufactured. The storage medium or media may be located on a machine that executes the machine-readable instructions or may be located at a remote site from which the machine-readable instructions can be downloaded over a network for execution.

[0056] It should be understood that computing system 100 is but one example of a computing system and that any other embodiment of a computing system may have more or fewer components than those shown, may combine additional components not shown in the exemplary embodiment of Figure 5, and / or computing system 100 may have a different configuration or arrangement of the components shown in Figure 5. The various components shown in Figure 5 may be implemented in hardware, software, or a combination of both hardware and software, including one or more signal processing and / or application specific integrated circuits.

[0057] Furthermore, the acts of the processing methods described above may be implemented by executing one or more functional modules in an information processing device, such as a general-purpose processor, or an application-specific chip, e.g., an ASIC, an FPGA, a PLD, or other suitable device, and all such modules, combinations of such modules, and / or combinations with common hardware are within the scope of the present disclosure.

[0058] In light of the principles and exemplary embodiments described and illustrated herein, it will be recognized that the exemplary embodiments may be modified in arrangement and detail without departing from such principles. The foregoing discussion has focused on specific embodiments, but other configurations are contemplated. In particular, even if expressions such as "in one embodiment" are used herein, these expressions are intended to generally refer to possible embodiments and are not intended to limit the disclosure to a specific embodiment configuration. As used herein, these terms may refer to the same or different embodiments that can be combined with other embodiments. In principle, any embodiment referred to herein can be freely combined with any one or more of the other embodiments referred to herein, and any number of features of different embodiments can be combined with each other unless otherwise specified. Although several examples have been described in detail above, those skilled in the art will easily understand that many modifications are possible within the scope of the described examples. Therefore, all such modifications are intended to be included within the scope of the present disclosure as defined in the following claims.

Claims

1. 1. A method for generating an image of a human face predicted to have at least one altered perceptual feature, comprising: encoding an image of a human face into a vector to generate an input image vector; applying the scaled direction vector to the input image vector to generate a modified input image vector; and converting the modified input image vector into a transformed image of a real human face; The method, wherein the scaled directional vector is determined using an artificial neural network trained to associate human faces with assessments of perceptual features to determine unit directional vectors, the unit directional vectors being based on at least one selected perceptual feature, and an amplitude for scaling the unit directional vectors being determined by limits of facial feature change obtainable with known treatments.

2. 2. The method of claim 1, wherein the unit direction vector is determined by selecting a highest rated human face rating image and a lowest rated human face rating image from among a plurality of human face rating images, the evaluation being performed by the artificial neural network for the at least one selected perceptual feature, and determining a direction of the unit direction vector from a vector representation of the selected highest rated image and a vector representation of the selected lowest rated image.

3. 3. The method of claim 2, wherein the evaluation images of human faces include randomly generated artificial human face images, the randomly generated artificial human face images being generated by a generative adversarial network.

4. 10. The method of claim 1, further comprising: using the input image and the deformed image to automatically generate at least one recommended treatment to make the human face from which the input image was created most similar to the deformed image.

5. The automatically generating the at least one recommended treatment includes: inputting the input image and the deformed image into an artificial neural network trained for change detection to detect particular regions in the input image that have changed; comparing the detected altered regions to a mask of facial regions known to be affected by a particular treatment; The method of claim 4 , further comprising: if an intersection of at least one of the detected change regions with at least one of the facial region masks is greater than a predefined threshold, the particular treatment is added to a suggested treatment list.

6. The method of claim 5 , wherein the change regions are detected by computing a difference image between the image of the actual human face and the deformed image.

7. The method of claim 5 , wherein the change regions are detected by inputting the image of the actual human face and the deformed image into an artificial neural network trained in change detection.

8. 5. The method of claim 4, wherein the automatically generating the at least one recommended treatment comprises inputting the input image and the transformed image into an artificial neural network that has been trained by inputting pre-treatment and post-treatment images of an actual human face that has received at least one treatment.

9. 5. The method of claim 4, wherein the automatically generating the at least one recommended treatment comprises: comparing a directional vector of the input image with reference to the deformed image with a directional vector of each of a plurality of treatments; and selecting one of the plurality of treatments that most closely matches the directional vector of the input image with reference to the deformed image.

10. The method of claim 1 , wherein the encoded image is an image of an actual human face.

11. A computer program stored on a non-transitory computer readable medium, the program causing a programmable computer to: encoding an image of a human face into a vector to generate an input image vector; applying the scaled direction vector to the input image vector to generate a modified input image vector; and Transforming the modified input image vector into a transformed image of a real human face; [0023] comprising logic operable to cause execution of actions including: The scaled directional vector is determined by training a first artificial neural network that determines a change in appearance of a human facial image corresponding to a change in the perceptual feature to determine a directional vector based on at least one selected perceptual feature, and an amplitude for scaling the directional vector is determined by limits of change in the facial feature obtained by known treatments.

12. 12. The computer program product of claim 11, wherein the unit direction vector is determined by selecting a highest rated human face evaluation image and a lowest rated human face evaluation image from among a plurality of human face evaluation images, the evaluation being performed by the artificial neural network for the at least one selected perceptual feature, and a direction of the unit direction vector is determined from a vector representation of the selected highest rated image and a vector representation of the selected lowest rated image.

13. 13. The computer program product of claim 12, wherein the evaluation images of human faces include randomly generated images of artificial human faces, the randomly generated images of artificial human faces being generated by a generative adversarial network.

14. 12. The computer program product of claim 11, further comprising: using the input image and the deformed image to automatically generate at least one recommended treatment to make the human face from which the input image was created most similar to the deformed image.

15. The automatically generating the at least one recommended treatment includes: inputting the input image and the deformed image into an artificial neural network trained for change detection to detect particular regions in the input image that have changed; comparing the detected altered regions to a mask of facial regions known to be affected by a particular treatment; and adding the particular treatment to a suggested treatment list if an intersection of at least one of the detected change regions with at least one of the facial region masks is greater than a predefined threshold.

16. The computer program product of claim 15 , wherein the changed regions are detected by computing a difference image between the image of the actual human face and the deformed image.

17. 16. The method of claim 15, wherein the change regions are detected by inputting the image of the actual human face and the deformed image into an artificial neural network trained in change detection. The computer program product described herein.

18. 12. The computer program product of claim 11, wherein the automatically generating the at least one recommended treatment comprises inputting the input image and the transformed image into an artificial neural network trained by inputting pre-treatment and post-treatment images of an actual human face that has undergone at least one treatment.

19. 12. The computer program product of claim 11, wherein the automatically generating the at least one recommended treatment comprises: comparing a directional vector of the input image with reference to the deformed image with a directional vector of each of a plurality of treatments; and selecting one of the plurality of treatments that most closely matches the directional vector of the input image with reference to the deformed image.

20. The computer program product of claim 1 , wherein the encoding is an actual human face.

Images