Papers by rodolphe korichi

Skin Research and Technology, Nov 1, 2000
BACKGROUND/AIMS: The different properties and visual effects of lipstick have been studied by ima... more BACKGROUND/AIMS: The different properties and visual effects of lipstick have been studied by image analysis directly on volunteers. METHODS: After controlling the volunteer's position mechanically using an ophthalmic table and visually using an acquirement mask, which is an indicator of luminance and guide marks, we carried out video colour images of the make-up area. From these images, we quantified the colour, gloss, covering power, long-lasting effect and streakiness, using computer science programs. RESULTS/CONCLUSION: Quantitative colorimetric assessment requires the transformation of the RGB components obtained by a video colour camera into CIELAB colorimetric space. The expression of each coordinate of the L*a*b* space according to R,G,B was carried out by a statistical method of polynomial approximations. A study, using 24 colour images extracted from a Pantone(R) palette, showed a very good correlation with a Minolta Colorimeter(R) CR 300. The colour assessment on volunteers required a segmentation method by maximizing the entropy. The aim was to separate the colour information sent back by the skin to the make-up area. It was very useful to precisely delimit the contour between the skin and the product in the case of almost identical colours and to evaluate the streakiness. From this colour segmentation, an algorithm was studied to search for the shades most represented in the overall colour of the make-up area. The capacity to replicate what the consumer perceives of the make-up product, to carry out studies without having any contact with the skin surface, and the constant improvement of software and video acquirement systems all make video imaging a very useful tool in the quantitative assessment of the properties and visual effects of a make-up product.

Skin Research and Technology, Nov 1, 2006
The individual perception of the radiance is difficult to define and quantify because it often in... more The individual perception of the radiance is difficult to define and quantify because it often includes physiological and psychological attributes. Although if they are differences in the perception of the skin radiance, dermatologists and experts in cosmetology consider that the complexion must reflect the general health of an individual. The aim of this work was to determine existing conditions of skin radiance via digital image analysis, consumer's perception and experts' evaluation, and propose a mathematical model to quantify the global radiance phenomenon. A round table meeting was organized in order to have free discussions about skin radiance. The qualitative assessment of skin radiance was carried out using a self-assessment questionnaire constructed according to the information obtained from the round table meeting. A simplified version of this questionnaire was proposed to experts. One hundred female subjects belonging to three different age groups (20-30, 30-40, 40-50 and >50 years old) participated in the study. Facial images were taken using a video imaging with visual control system to ensure consistency of images among volunteers. From these images, algorithms based on the principle of polynomial approximation, segmentation by maximalization of the entropy and the Logarithmic Image Processing model were required to extract skin radiance parameters. Multiple regression analysis was used to establish, via consumer's perception of the skin radiance and experts' evaluation studies, the degree to which each of the independent parameters contributes to the skin radiance perception. All the items of the self-assessment questionnaire explained 74% of the variation of the skin radiance. Data from experts' analysis explained 87% of the variation of skin radiance. In the two cases, skin radiance appeared to be independent of age and the presence of wrinkles. From the image analysis, we observed that the skin surface parameters explained 20% of the variation of the skin radiance expressed by experts and 24% of the variation of the skin radiance expressed by volunteers. As a result, about 80% of the variation is not explained by the instrumental data. If we combine experts' analysis and instrumental data, we obtain a mathematical model that explains 92% of the variation of the skin radiance. The questionnaire constructed from this brainstorming session allowed volunteers and experts to judge their perception of skin radiance more precisely. Video imaging appears as an interesting method to quantify visual properties of the skin and to visualize what the consumer perceives of skin radiance. However, the mathematical model proposed from the skin surface parameters analysis does not totally explain the global perception of skin radiance. In the future, it will be necessary to develop new data-processing programs to quantify subjective parameters in order to validate our mathematical model.
arXiv (Cornell University), Feb 24, 2020
The lack of transparency of neural networks stays a major break for their use. The Layer-wise Rel... more The lack of transparency of neural networks stays a major break for their use. The Layer-wise Relevance Propagation technique builds heat-maps representing the relevance of each input in the model's decision. The relevance spreads backward from the last to the first layer of the Deep Neural Network. Layer-wise Relevance Propagation does not manage normalization layers, in this work we suggest a method to include normalization layers. Specifically, we build an equivalent network fusing normalization layers and convolutional or fully connected layers. Heat-maps obtained with our method on MNIST and CIFAR-10 data-sets are more accurate for convolutional layers. Our study also prevents from using Layer-wise Relevance Propagation with networks including a combination of connected layers and normalization layer.
ArXiv, 2020
The lack of transparency of neural networks stays a major break for their use. The Layerwise Rele... more The lack of transparency of neural networks stays a major break for their use. The Layerwise Relevance Propagation technique builds heat-maps representing the relevance of each input in the model s decision. The relevance spreads backward from the last to the first layer of the Deep Neural Network. Layer-wise Relevance Propagation does not manage normalization layers, in this work we suggest a method to include normalization layers. Specifically, we build an equivalent network fusing normalization layers and convolutional or fully connected layers. Heatmaps obtained with our method on MNIST and CIFAR 10 datasets are more accurate for convolutional layers. Our study also prevents from using Layerwise Relevance Propagation with networks including a combination of connected layers and normalization layer.

A major issue in machine learning is to select the best hyperparameters of a predictive model wit... more A major issue in machine learning is to select the best hyperparameters of a predictive model without over-fitting. In this paper, we propose to study through a principled way the hyperparameter optimization in a neural network designed for a classification problem on cosmetic formulation data. Specifically, we propose to make use of Bayesian optimization (BO) to automatically choose the next hyperparameter set to try based on previous observations and a surrogate function of the model. This BO-based hyperparameter selection method is compared to the popular grid search method. A 2-hidden-layer fully-connected neural network is trained on cosmetic formulation data. Extensive experiments show that hyperparameters found with Bayesian optimization outperform the grid search hyperparameter set. Moreover, Bayesian optimization only needs 80 evaluations while 140 are used for the grid search.
The present invention relates to a method for characterizing the distribution of volumes of the s... more The present invention relates to a method for characterizing the distribution of volumes of the surface of the skin of an individual, particularly finding application in the evaluation of the effect of firming cosmetic products. This method uses the contours and gradients of slopes calculated from a three-dimensional representation of the surface of the skin.
Médecine du Sommeil, 2018

Sleep, 2019
Introduction: Sleep deprivation has a strong impact on facial appearance with visible associated ... more Introduction: Sleep deprivation has a strong impact on facial appearance with visible associated signs of fatigue but the impacts on skin biophysical features still remains unknown. Methods: The present study investigates the impact of two consecutive nights with sleep restriction to 3 hours per night on facial skin, by using noninvasive instrumental methods for sebum (sebumeter), hydration (corneometer CM 825/Courage & Khazaka), trans-epidermal water loss (tewameter TM 210), biomechanical properties (cutometer MPA 580), pH (PH 900), desquamation (D-squameter and image analysis) quantification and image analysis (ColorFace/Newtone Technologies). Twenty-four healthy Caucasian women aged 30-55 were selected for their well-sleeping with Epworth scale, Pittsburg quality index and homogeneity of their circadian typology with Horne and Ostberg questionnaire. Sleep deprivation was controlled by actimetry, and a sleep diary. Biophysical measurements and image of the face were taken in the morning and the evening the day before and the day after the sleep deprivation. Results: We observed that sebum and PIE were more intense in the afternoon than in the morning without any impact of sleep deprivation on these parameters. On the contrary, hydration of the skin surface, pH and viscoelastic properties of the skin e.g. extensibility (Uf), elasticity (Ue), and delayed extensibility (Uv) as well as well single desquamating corneocytes (invisible desquamation) and radiance were significantly reduced after sleep deprivation. Conclusion: This study identifies several important skin facial parameters significantly affected by acute sleep deprivation e.g. hydration, acidity, desquamation homogeneity, viscoelastic properties and complexion luminosity and highlights the importance of the sleep quality (or length) in physiological skin homeostasis, aesthetic and comfort. Support (If Any

Médecine du Sommeil, 2019
Méthodes Enquête transversale mono-centrique menée du 1 er janvier 2016 au 30 septembre 2017. Tou... more Méthodes Enquête transversale mono-centrique menée du 1 er janvier 2016 au 30 septembre 2017. Tous les patients hospitalisés pour suspicion du syndrome d'apnées du sommeil ont bénéficié d'une oxymétrie nocturne et d'une polygraphie ventilatoire. Les logiciels Excel 2010 et SPSS 21.0 ont permis d'analyser les données. Nous avons déterminé la sensibilité, la spécificité, la valeur prédictive positive, la valeur prédictive négative et la courbe ROC (p < 0,05). Résultats Un total de 201 patients d'âge moyen de 64,6 ± 11,8 ans, avec une prédominance masculine (55 %) et en majorité obèses (IMC moyen de 32 kg/m 2) ont été inclus. Le ronflement (71 %), la somnolence diurne (71 %) et la fatigue diurne constituent les principaux signes d'appels. La sensibilité et la spécificité de l'oxymétrie nocturne sont respectivement de 87 et de 85 % avec une courbe ROC à 0,75. Conclusion L'oxymétrie nocturne quoi que non encore formellement recommandée, est un examen non invasif, peu coûteux et facile d'accès ayant une sensibilité et une spécificité élevée dans le diagnostic du syndrome d'apnées du sommeil. Déclaration de liens d'intérêts Les auteurs déclarent ne pas avoir de liens d'intérêts.
Revue de Chirurgie Orthopédique et Traumatologique, 2016
Journal of Investigative Dermatology, 2017

Cosmetics are part of behavioral and psychosocial dynamics. In this theoretical framework, a firs... more Cosmetics are part of behavioral and psychosocial dynamics. In this theoretical framework, a first series of experiments showed two very distinct and opposed psycho-behavioral profiles called « camouflage » and « seduction ». In a second series of experiments, the two profiles were linked to specific variations in the type of skin (visual quality), to the asymmetry of some facial features, as well as to the use of colors in making-up. Finally, combining the different ethological and psycho psychological approaches, we showed that making up allows to reduce emotional reactions in a situation of social stress, mainly among women belonging to the « camouflage » profile. In conclusion, making-up appears as a holistic technique which allows to manipulate not only the aesthetics of a face, but also the emotional responses linked to the image of oneself and the interactions with a social environment.
Uploads
Papers by rodolphe korichi