Custom Color Clothing Recommendation System and Method Based on Customer Skin Tone

Abstract

The present invention relates to a system and method for classifying personal color (skin tone) by analyzing a customer's skin image and recommending personalized colored clothing by searching a clothing color database based on the results. After correcting color distortion caused by the shooting environment (lighting, white balance), an optimal color palette for each user is determined using a machine learning-based personal color classification model, and a recommendation list is generated by matching this with clothing color information. Furthermore, the accuracy of recommendations is enhanced by continuously reflecting the user's purchase history and preferences, and the invention is widely applicable in the fields of fashion and e-commerce.

Description

Technology Field

[0001] The present invention relates to user-customized services in the fashion and beauty fields, and more specifically, to a customized color clothing recommendation system and method that analyzes a customer's skin tone (personal color) and recommends clothing of suitable colors based on the results. More specifically, it belongs to the field of technology that provides personalized fashion color matching using image processing, machine learning algorithms, database integration, etc. Background Technology

[0003] Recently, personalized recommendation technologies are being actively developed and utilized in online and mobile environments. For example, product recommendation systems based on customers' past purchase history and preferred categories have become commonplace on e-commerce platforms. However, in cases involving highly subjective elements, such as clothing color, where suitability varies significantly depending on individual physical characteristics (like skin tone), it is difficult to provide accurate and highly satisfactory results using existing simple recommendation algorithms.

[0004] Previously, personal color consulting was often performed manually by experts or limited to simply extracting RGB values ​​to classify skin tones. Additionally, there is a problem where actual skin tone measurements are distorted by external factors such as lighting and shooting environments, requiring a separate algorithmic approach to correct this.

[0006] Accordingly, there is an increasing demand for systems that more precisely measure and analyze skin tone and link the results with a fashion clothing color database to recommend items that users are likely to prefer. The problem to be solved

[0010] To solve the problems of the prior art described above, the present invention presents a **“Customer Skin Tone-Based Customized Color Clothing Recommendation System and Method”** that can accurately measure and classify a customer’s skin tone and efficiently provide personalized color clothing recommendations based on the classification results.

[0012] Specifically, the present invention aims to solve the following problems.

[0014] It provides technology that can stably extract skin tone information by minimizing the influence of the shooting environment (lighting, background, etc.).

[0015] It systematically classifies customers' skin tone types and performs color recommendations by automatically mapping information within the clothing color database.

[0016] We provide customized filtering features to reflect additional customer preferences (body type, style, price range, etc.).

[0017] By analyzing feedback data (preferences and purchase history) regarding recommendation results, the algorithm can be continuously refined. means of solving the problem

[0021] 1. Technology forming the background of the invention

[0022] 1.1. Current Status of Development of Personalized Recommendation Technology

[0023] As the e-commerce market grows, personalized recommendation technology is being applied across various fields. Notably, customized product recommendation systems that consider individual tastes and styles have been actively researched and developed in the fashion industry, in addition to recommendations for movies, music, and books. These recommendation technologies are evolving to learn user behavior patterns and preferences by utilizing machine learning and deep learning techniques, as well as large-scale user data.

[0025] Collaborative Filtering: A method of identifying user groups with similar tastes and recommending products preferred by those users.

[0026] Content-Based Filtering: A method of analyzing the attributes (color, material, author, genre, etc.) of a specific product (or content) and recommending other products with attributes similar to the user's preferences.

[0027] Hybrid method: A method that improves recommendation accuracy by combining collaborative filtering and content-based filtering.

[0028] Conventional clothing recommendation systems primarily focus on collecting and analyzing users' click history, purchase history, and shopping cart data to recommend similar products or trendy items. However, visual and aesthetic aspects, such as clothing color, are subject to distinct personal preferences; furthermore, the degree of "fit" varies significantly depending on physical characteristics (skin tone, body type, image). In this context, there is a problem in that it is difficult to provide accurate color recommendations based solely on preference data.

[0030] 1.2. Personal Color (Skin Tone) Analysis Technology

[0031] Meanwhile, **Personal Color** refers to the color tone characteristics inferred from an individual's skin tone, hair color, iris color, etc., and the color palette that suits them. In the beauty and fashion industries, various classification methods have been utilized, such as the four seasons (SPRING, SUMMER, AUTUMN, WINTER) or 12 seasons (detailed tone classification) models. Previously, experts (personal colorists) determined individual customers' tones by manually comparing color charts and applying test strips or makeup.

[0033] However, these traditional methods are time-consuming and rely on subjective judgment, making it difficult to obtain consistent results, and are nearly impossible to apply in large-scale online environments. Accordingly, research is being conducted on a technology that combines computer vision (image processing) technology and machine learning to analyze face images uploaded by users, extract quantitative skin tone color values ​​after correcting lighting and background, and then algorithmically map them to personal color types.

[0035] 1.3. Limitations of Existing Technology

[0036] Lighting and White Balance Issues: Photos taken by general users with mobile phone cameras often show distorted skin tones due to lighting, shooting angles, and background colors. If these are not properly corrected, inaccurate color analysis results will occur.

[0037] Limitations of simple RGB extraction: The RGB (Red, Green, Blue) method alone makes it difficult to accurately match the human subjective visual system, and values ​​fluctuate significantly when illumination or white balance conditions change. Consequently, color space conversion using tools like CIELAB is necessary, and sophisticated algorithms are required.

[0038] Lack of reliability in personal color classification models: Since personal color must reflect not only individual characteristics (race, skin type, hair color, innate pigmentation) but also subjective preferences to some extent, it is difficult to achieve actual satisfaction with simple rule-based classification alone.

[0039] Difficulties in Real-Time Large-Scale Processing: As an online service where multiple users simultaneously upload images and request color analysis, it is necessary to ensure both fast response speeds and accuracy.

[0040] To solve the above problems, sophisticated image preprocessing algorithms, machine learning / deep learning-based personal color classification models, and integration technology with clothing color tagging databases are required. In addition, a method of continuously improving the system through user feedback (preference / purchase history) learning functions is required. Effects of the invention

[0044] According to the present invention, a customer can obtain highly accurate skin tone analysis results by simply providing or capturing an image of their skin. Furthermore, by automatically linking these analysis results with a fashion clothing color database, the customer can receive real-time recommendations for clothing optimized for their personal color. In addition, the following specific effects are achieved.

[0046] Improved accuracy: Reduces measurement errors and increases the reliability of the skin tone classification model through lighting correction, color space conversion, etc.

[0047] Enhanced Personalization: Satisfaction with recommendation results increases by considering not only skin tone but also the customer's purchase history and style preferences.

[0048] System Efficiency: Since automated algorithms can simultaneously provide recommendation services to a large number of users, the scope of commercial and practical application expands.

[0049] Continuous Learning: By receiving feedback on the user's selection and purchase history after a recommendation, the machine learning model can be improved, thereby increasing long-term recommendation accuracy. Brief explanation of the drawing

[0052] Figure 1 shows customer skin tone and purchasing tendencies. Specific details for implementing the invention

[0053] (1) Overall System Overview

[0054] The system according to the present invention is largely composed of a customer terminal (smartphone, webcam, etc.), a server (skin tone analysis and recommendation algorithm), and a clothing database.

[0056] Customer Terminal: The customer uploads an image of their skin or takes a photo in real-time and transmits it to the server.

[0057] Server: After receiving the transmitted image, extracts and classifies skin tones through correction and analysis algorithms.

[0058] Clothing DB: Clothing of various colors is stored along with attributes (color code, style, price, etc.), and the server retrieves a list of clothing corresponding to the color range that matches the customer's skin tone.

[0059] (2) Skin tone analysis procedure

[0060] Image Preprocessing

[0062] Lighting and White Balance Correction: Minimizes skin color distortion by utilizing the uploaded image's metadata (illumination, color temperature, etc.) and background information.

[0063] Skin Region Segmentation: Extract the Region of Interest (ROI) using a face recognition algorithm or a skin pixel classification algorithm.

[0064] Color space conversion (RGB → CIELAB, etc.): Converts to a color space closer to human visual characteristics to calculate the average skin tone value.

[0065] Personal Color Classification

[0067] The collected skin tone values ​​are input into a personal color classification model (e.g., Spring / Summer / Autumn / Winter, further subdivision) or a machine learning-based prediction model.

[0068] The classification model may be trained on a previously labeled skin tone-personal color dataset and may reflect the user's race, regional characteristics, etc.

[0069] (3) Clothing recommendation algorithm

[0070] Color Matching

[0072] Determine a predefined recommended color palette (e.g., HEX, PANTONE, LAB range, etc.) based on the classified skin tone type.

[0073] Filter the clothing database based on colors belonging to the corresponding palette range or similarity (ΔE value, etc.).

[0074] Personalized filtering

[0076] To reflect the customer's body type, style preferences, and budget range, the final recommendation ranking is determined by combining collaborative filtering (past purchase history) or content-based filtering (product metadata).

[0077] Feedback learning

[0079] It stores the results of the customer's actual selection and purchase from the recommendation list (preference / non-preference).

[0080] Machine learning / deep learning algorithms retrain this data to improve future recommendation accuracy.

Claims

Claim 1 (1) Overview of the Overall System The system according to the present invention is largely composed of a customer terminal (smartphone, webcam, etc.), a server (skin tone analysis and recommendation algorithm), and a clothing database. Customer Terminal: The customer uploads an image of their skin or takes a picture in real-time and transmits it to the server. Server: After receiving the transmitted image, the server extracts and classifies the skin tone through a correction and analysis algorithm. Clothing DB: Clothing of various colors is stored along with attributes (color code, style, price, etc.), and the server retrieves a list of clothing corresponding to the color range that matches the customer's skin tone. (2) Skin Tone Analysis Procedure Image Preprocessing Lighting and White Balance Correction: Skin color distortion is minimized by utilizing the metadata (illumination, color temperature, etc.) and background information of the uploaded image. Skin Region Separation: Regions of Interest (ROI) are extracted using a face recognition algorithm or a skin pixel classification algorithm. Color Space Conversion (RGB → CIELAB, etc.): The average skin tone value is calculated by converting to a color space closer to human visual characteristics. Personal Color Classification The collected skin tone values ​​are processed using a personal color classification model (e.g., The classification is input into a Spring / Summer / Autumn / Winter (additional subdivision) or a machine learning-based prediction model. The classification model may be trained on a previously labeled skin tone-personal color dataset and may reflect the user's race, regional characteristics, etc. (3) Clothing Recommendation Algorithm Color Matching A predefined recommended color palette (e.g., HEX, PANTONE, LAB range, etc.) is determined based on the classified skin tone type. The clothing database is filtered based on colors belonging to the corresponding palette range or similarity (ΔE value, etc.). Personalized Filtering To reflect the customer's body type, style preference, and budget price range, the final recommendation ranking is determined by combining collaborative filtering (past purchase history) or content-based filtering (product metadata). Feedback Learning The results of the customer's actual selection and purchase from the recommendation list (preference / non-preference) are stored.Machine learning / deep learning algorithms retrain this data to improve future recommendation accuracy. Claim 2. A method according to Claim 1, wherein in step (b), the lighting and white balance correction processing further includes the step of correcting color temperature and brightness based on metadata to minimize image distortion according to the shooting environment, and converting the color space from RGB to CIELAB before extracting the skin tone color value. Claim 3. A method according to Claim 1 or 2, wherein the personal color classification model includes a machine learning algorithm that learns user profiles (race, age, skin characteristics) and a dataset labeled by a personal color expert, and automatically determines the skin tone type for each customer. Claim 4. A method according to any one of Claims 1 to 3, characterized by further including the step of applying a collaborative filtering or content-based filtering algorithm to additionally consider the customer's style preference, body shape information, price range preference, etc. when generating the recommendation list. Claim 5. A method according to any one of Claims 1 to 4, further comprising the step of improving recommendation accuracy by collecting the customer's selection or purchase history after providing the recommendation list and retraining the personal color classification model and clothing recommendation algorithm with it.

Images