WO2018109699A1

WO2018109699A1 - Systems and methods for matching garments

Info

Publication number: WO2018109699A1
Application number: PCT/IB2017/057926
Authority: WO
Inventors: Maciej Szamotulski
Original assignee: Maciej Szamotulski
Priority date: 2016-12-14
Filing date: 2017-12-14
Publication date: 2018-06-21
Also published as: GB201621229D0

Abstract

A system for matching garment items, the system comprising a database comprising at least one user profile, wherein at least one user profile comprises two or more user profile parameters, the database further comprising a plurality of garment sets, each garment set comprising two or more garment items (including, but not limited to new and second-hand garments) from said garment items (available via a network or marketplace community of users or commercial seller or sellers), the system further comprising a processor for mapping a correspondence between the plurality of garment sets respectively to the at least one user profile.

Description

Systems and Methods for Matching Garments

Background

The present invention relates generally to systems and methods for matching garments and related machine learning methods and tools. More specifically, the present disclosure relates to an automated fashion management system based on clothes recognition, user profiling and clothes matching, and style ("look") recommendation, amongst others.

Prior Art

Clothes matching and organising is a common activity which can be carried out using clothes from personal wardrobes, or whilst shopping in physical stores or online. A common practice in a physical store is to search the inventory for items of interest, select a few for comparison, and try them on. Clothes are usually matched according to how well they fit physically, and also how well they fit an image or "look" that one might want others to perceive. Usually, shoppers not only check whether garments fit their bodies, but also whether they fit their style, and their wardrobes (other garments). Sought-after styles and premium brand items are often expensive.

It is common for purchased clothes to accumulate in a person's wardrobe over time and be forgotten eventually or be considered out of fashion by the owner. Many people then find themselves in situations where they believe they have "nothing to wear" or match from their wardrobe which is full of clothes, which are now underutilised assets, or investment. This leads the person to purchase more clothes which may or may not match the items already existing in their wardrobe, or to purchase full outfits leading to unnecessary expense. Typically, fashion and clothes shopping decisions are driven by a person's goals for self- expression within their social context at the time, which is determined by many complex and subtle factors, including budget/affordability. A particular article of clothing may present a different aesthetic depending on how it is coordinated with other articles. A particular style may be considered "cool" or "inappropriate" in different occasions and contexts. In addition, a particular item may go "in" and "out" of fashion over time and among different demographic segments.

Known activities for matching garments are thus inefficient and time consuming. Further, there is no affordable technology at present that would replace a professional wardrobe organiser or human curator, with the same effectiveness. There are no existing systems to quickly and effectively match shopper styles to disposable budgets, let alone quickly and efficiently identifying more affordable replacement used items fitting shopper's style, and budget. There is therefore a need for systems and methods which address at least these problems. Summary

In a first independent aspect, there is provided a system for matching garment items, the system comprising a database comprising at least one user profile, wherein at least one user profile comprises two or more user profile parameters, the database further comprising a plurality of garment sets, each garment set comprising two or more garment items (including, but not limited to new and second-hand garments) from said garment items (available via a network or marketplace community of users or commercial seller or sellers), the system further comprising a processor for mapping a correspondence between the plurality of garment sets respectively to the at least one user profile.

A "garment set" corresponds to a look, and individual items of clothing comprising such a look, be it new or used items or a combination thereof and such different looks may be correlated to user parameters such as "stylity" or "on-trend-iness" or "disposable budget" as will be described below. In contrast to existing activities, the correlation is achieved by the system more efficiently, in a quantifiable manner to ensure the best fit to the user profile. Mapping a correspondence may be done for example using a correspondence matrix between individual user items and user profile parameters. Several suitable algorithms are envisaged. Preferably, the mapping takes into account a consistency between a look and the user profiling parameters. In a second, independent aspect, there is provided a method for matching garment items to a user profile, the method comprising the steps of: providing a plurality of garment items;

providing at least one user profile, wherein at least one user profile comprises two or more user profile parameters,

providing a plurality of garment sets, each garment set comprising two or more garment items from said plurality of garment items, and mapping a correspondence between the plurality of garment sets respectively to the at least one user profile. Dependent aspects, representing preferred embodiments of the invention, are set out in the dependent claims.

Preferably, the processor is configured to assign a first garment set to the at least one user profile. This effectively recommends the look to the user.

Preferably, the processor is further configured to map a correspondence between the first garment set and a further plurality of garment items. Mapping this correspondence may be done using the same or similar algorithms to those used for mapping a correspondence between the plurality of garment sets respectively. This effectively recommends a particular item to the user. The recommended (assigned) garment items may be from the same database, or other databases.

The system and methods described here may be used in many applications, including mobile applications loaded on portable devices such as a smart phone or tablet. Further applications are envisaged, whereby the system is included in a wardrobe organiser or virtual shop assistant system. The virtual shop assistant system may be a mobile application for example. A wardrobe organiser application is envisaged, whereby the system recommends acquiring a particular item from another database (e.g. from an online shop) or recommends disposing of unused items which have not been used in a long time for example, or which cannot be matched to a satisfactory criteria to other items.

Brief Description of the Drawings The disclosure will now be described with reference to and as illustrated by the accompanying drawings in which:

Figure 1 shows a system according to an aspect of the disclosure; Figure 2A shows an example of "look" itemisation; Figure 2B shows an example of user profiling; Figure 2C shows another example of user profiling;

Figure 3 shows an example of data sets used for machine learning (look training);

Figure 4 shows an example of data (look) classifiers; Figure 5 shows an example of user-to-look mapping (and look retrieval);

Figure 6 shows an example of a networking system including a "virtual closet" fashion repository; Figure 7 shows an example of garment matching for a first user type group;

Figure 8 shows an example of garment matching for a second user type group; Figure 9 shows an example of garment matching for a third user type group ("specific searchers buyers");

Figure 10 shows a price matching mechanism whereby shoppers budget determines the combination of new and second-hand garments the shopper will see in the recommendation;

Figure 1 1 shows a look retrieval for a use case/scenario;

Figure 12 shows another example of user-to-look mapping (and look retrieval);

Figure 13 shows screenshots of an application according to aspects of the invention; Figure 14 shows further screenshots of an application according to aspects of the invention; Figure 15 is an example look map based on variables for trend followers versus stylity; Figure 16 is an example look map based on variables for formality versus stylity; Figures 17 A and 17B are examples of journey plot in the look-space;

Figure 18 shows an example of k-means clustering with k=6;

Figures 19A and 19B show plot overlays of examples of looks where the heights (yellow areas Y) indicate highest incidence of data points (or particular fashion styles) in the space; and

Figure 20 shows an alternative discovery journey plot based on stylistic similarity. Detailed Description

Figure 1 shows an exemplary system that facilitates clothes-recognition, user profiling and clothes matching. A computer system or mobile device 802 includes a processor 804, a memory 806, and a storage device 808. The computer system 802 is coupled to a display 801 and a camera 803. It will be appreciated that the computer system, display and camera may be integrated in a smart phone or tablet for example.

The storage device 808 stores code for an operating system 816, as well as locally cashed code of applications 820 and 822 including fashion and user data sets. In this example, applications correspond to a "look engine" and a "virtual closet" repository as will be described below. It will be appreciated that applications 818, 820 and 822 may be combined as a single package, and stored on cloud devices (server). The virtual closet 822 uses a shared database which may be stored on a cloud computing system for example and is accessible over an internet network 900. Also included in the storage device 808 in this example are clothes-recognition applications 818 with image recognition capability. During operation, applications are loaded into memory 806.

The computer system 802 is coupled to the internet network 900 to which a seller 950 (or other user) is also connected. It will be appreciated that the system or systems connected to the network may be accessed by more than one user accessing a similar system, e.g. the users having downloaded the same application(s), and more than one seller. Connections with the network can be wired or wireless. When the processor 804 executes the corresponding code stored in memory 806, the processor 804 may perform itemization, user profiling, machine learning, image analysis to the images captured by camera 803, and networking with other users/seller as will be described below. The display 801 may display user interfaces with the applications. The display 801 may also be a touch screen which allows users to interact with the applications.

The camera 803 may be used by users to scan in items of clothes for example. The items may then be uploaded to the "look engine" and virtual closet applications. The virtual closet stores data sets including digital images and associated description and user ID for example. With reference to Figure 2A, "look" types (also referred to as a "look" or "style") 10 are defined as a combination of garments 1 1, 12, for example a top such as a blouse plus a bottom, such as a skirt. An application 820 referred to as the "look engine" has an itemisation module configured to "itemise" the look, that is, to identify and extract items types (e.g. blouse, skirt) and other classifiers (e.g. shape, pattern) which in combination form the "look". The application then enables creating a plurality of combinations for the same look using different but similar items in colour and/or shape/pattern. For example, a "blouse + skirt" look can include a black blouse + black skirt or a grey blouse + black skirt or a grey blouse + grey skirt etc. User input to "curate" the looks is also enabled. A plurality of "looks" forms fashion data which may be then "trained" as will be described below in the "Machine Learning/Training/Discovery Mechanisms" section.

With reference to Figure 2B, the look engine application comprises a module 821 referred to as the "profiling" module 821 configured to create a user profile based on several descriptors such as: fashion, utility or personality each descriptor being defined by one or more parameters. For example, parameters for fashion include: age, fitness, body type, best fit; parameters for utility include season (e.g. winter, summer) and scenario (e.g. business lunch, Friday night drinks, wedding) parameters for personality include: public information and signals (relating to user information available on social media), survey data, and behavioural data such as purchasing history, usage history, visits to websites, contact centre information, friends- all of which can define personality types. The user selects parameters from a list for example and can also make negative selections of items that do not fit them for example. The user profile therefore comprises a number of parameters defining one or more descriptors. A plurality of user profiles forms user data which may be then "trained" as will be described below.

Furthermore, to determine user fashion preference, a series of images and survey questions are supplied to the user, and the respective responses in turn allow the application to attribute specific weights in relation to given classifiers (descriptors), forming a synthetic (weight-based) numerical profile of user fashion choices across a set of fashion items, styles, and scenarios, and personalities, such that a dominant descriptor or descriptors including personality type, need/scenario, item preference as identified per each user. Figure 2C illustrates the creation of a synthetic user profile based on a list of descriptors (one per each column).

Creation of look types, and matching look types to user profiles may also be done using human input (fashion being hand curated) for example in the first instance. Preferably, the human curation will thereupon be superseded by Al machine learning and neural networks when a critical mass of users is ensured. Machine learning / Training/ Discovery Mechanisms

The training of the fashion data and user data sets is now described. With reference to Figure 3, the "look engine" 820 is configured, using machine learning techniques, to train the fashion and user data sets, which are data sources stored in the storage device 808, and in the "cloud". By training we mean evolving the data sets to improve the matching as well as taking into account that fashion and user preferences naturally evolve. The training in this example may be done in phases, which include: curated fashion which represents the learning phase, street fashion which represents the transfer phase, and continuous improvement which is a deployment phase.

For example, in the learning phase, fashion data is evolved by machine learning algorithms which are fed with well-matched pictures of consistent looks. In the transfer phase, the fashion data set is increased to include online content with additional look images representing additional look types, additional items etc.

Additionally, in the learning phase, user data is evolved by user survey data to determine for example, style preferences of various user personality types. In the deployment phase, the matching of personality traits with stylistic choices is continuously improved.

With reference to Figure 4 and Look Matrix 11, 11 represents the product of mapping look types 10 (represented by rows Look 1, Look 2, etc.) into classifiers (columns) such as garment types, shapes, colours, patterns, seasonality and other descriptors as per Figure 4.

The Look Matrix 11, allows us to match a particular user preferences based on his/her profiling (including his/her particular garment type, shape, utility and other preferences) against look types. A matrix is then called up for each individual user that effectively maps the correspondence between each look type (listed in each matrix row) and the user profiling parameters. The consistency between a look and the user profiling parameters is quantified in the matrix as a consistency match level. This is computed as a % of a perfect score whereas the perfect score would be achieved by the rarest of looks that fulfil upon each user criterion to 100% degree. This allows the algorithm to sort the looks based on consistency match level from most to least-matching. With reference to Figure 5, a method of user-to-look mapping or "look retrieval" (referred in this Figure as a "style") includes the following steps: 1. A "dominant descriptor" (such as personality, need, or style preference for example) is selected, for example via user interaction with the app, or on-screen choices by the user. That is the user may select from a user interface the most important criteria for the look, referred to as a "dominant descriptor";

2. Looks are matched using the matrix;

3. Looks are retrieved and displayed to the user in a particular order (recommendation sequence aka "journey").

In the example shown in Figure 5, the user asks for a particular scenario or use-case such as "/ want something to wear to drinks on Friday night", by selecting scenario as the most relevant descriptor (dominant descriptor being utility). The look engine 820 selects the look which is ranked highest in the matrix and displayed to the user.

A preferred machine learning algorithm run by the look engine is referred to as a "discovery" mechanism. The aim is that looks are curated by the look engine module, and accurately and efficiently matched to a user based on style preference, and shopping budget. This is advantageous to recommending looks based on body shape figure because every look can be transposed onto different body shapes.

A look is described by a pair of parameters such as "stylity" or "on-trendiness". Figures 15 shows data points representing looks which are plotted on a "trend follower" ("x" axis) scale versus "stylity" ("y" axis). The scale is from -100 to 100 in this example. Alternatively, the looks may be plotted on a "formality" versus "stylity" graph, as shown in Figure 16.

Using these parameters plotted in the "x" and "y" axes, and the aim of the discovery mechanism is to map distances between looks. This is done by calculating a distance between two looks (2 data points) in Euclidian space whereby proximity is a proxy for similarly and preference. Preferably, the system calculates all distances from one look to all looks that a user may have originally picked as preferred (e.g. during profiling). Based on the shortest distance calculated, the system can then select the closest look to the original look. The discovery mechanisms may map each look onto two most polarising variables (such as personality or style), assigning points based on how well the look matches those variables.

To arrive at a new look that is an effective match, the system may output a number of looks, in sequence. There is a "journey" therefore from the original look to the one which is ultimately a best fit, in an Euclidian space, mapped onto a log spiral in that space. This allows for granularity close to core user preferences, while providing a degree of novelty in the discovery, as well as surprise as users move away from the original "starting" look and arrive at a multitude of looks along the "journey".

For there to be a journey, there must be a route via points on the map (space) to journey from/to. Journeys can be wide, narrow, linear, fuzzy or else but they must cover an area. The larger the area that defines the range of a look selected by user (e.g. at user profiling, or algorithmically via choices made by the user), the more opportunity for the journey.

An example of a spiral journey plot is shown in Figure 17A, with the curve being represented by the following equations:

y(t) = r(i) sin(i) = ae^bt siii(t)

The curve represents a step in which a user sweeps the entire look-space, and the number of such curves (steps) may vary and may be controlled. Parameter "b" in the above equations controls the "tightness" of the spiral. For example, there may be a recommended interval of 90 degrees to define the sequence of matched looks which means that a point is selected every 90 degrees along the journey spiral. In other words, the journey can output a surprise look at each 90-degree turn.

When the user profile matches disparate looks, the recommended journey will be "jagged". To overcome this, data clustering may be achieved (by any suitable clustering algorithm) to smooth out the journey plot. Adjacent and neighbouring looks along the journey are clustered together to ensure homogeneity, or continuity of user experience. Clustering means grouping similar looks along the discovery journey plot. Figure 17B shows an example of clustering, with clusters indicated as (A), (B), (C), (D).

Initial cluster centres may be taken as clusters matching individual user profiles for example. The nearest look clusters are then grouped along the discovery path (journey spiral) and this may be done by a suitable algorithm such as k-means clustering (https://en.wikipedia.org/wiki/K-means clustering) for example. K-means clustering aims to partition n data points (looks in this case) into k clusters in which each data point belongs to the cluster with the nearest mean, serving as a prototype of the cluster. This results in a partitioning of the data space into Voronoi cells. The "centroid" or cluster centre is a data point which may be represented by a visual reference image. Preferably, this is displayed visually to a user interface as the most representative look of the cluster/group.

The process of clustering in the look-space acquires ("snaps") a well-matched look to the closest cluster. One data point (look) belongs only to one cluster. As an example, Figure 18 shows Voronoi cells for the process of k-means clustering with k=6. The discovery mechanism for the look engine described above calculates look proximity (distances between looks). It will be appreciated that other suitable discovery mechanisms may be run by the look engine.

For example, in an alternative, the look engine considers both look proximity as well as stylistic (behavioural) preferences as parameters. In this example, a style "demand" may be overlaid on the look-space map (e.g. as defined by x=stylity, y=ontrendiness parameters).

With reference to Figures 19A, 19B, a composite shaded area plot is an overlay of all looks (fashion styles) where the heights (yellow areas Y) indicate highest incidence of data points in the space— intersection of such styles. The peaks (Y) in the look space are the result of points of high incidence (high demand) "depositing" onto the image, indicating interest areas for most users, and likely styles users will choose during profiling, which are beginnings of their respective journeys. In this example, the elevated area/plateau spreads over a ½ of terrain with a peak at [20,20] and fanning out towards LHS with increasing Y spread. Such elevated areas indicate there may also be opportunities for not so obvious "vertical journeys" or "jumps" into other styles sharing common parameters.

A discovery mechanism which considers both proximity and stylistic similarity is now described, whereby one can induce user preference up into general style and match particular looks along discovery journey. With reference to the example in Figure 20, different styles are represented by different symbols. Users may "snap" to closest style neighbour (within the same style), but may also be transitioned onto a different style when sufficiently close. The point of when the journey moves on to the different style may be calculated, for example, with regard to the following:

1 ) Journeys that happen between points "3" and "4" are exciting for the user, but may be tiring when abused.

2) Such journeys have a lot of surprise potential.

3) Such transitions should this happen too early in the journey.

4) There is clustering potential along such journey points as "3" and "4".

Once users are transitioned onto a different look, the journey can continue based taking into account regular stylity parameters. That is, the journey may be adjusted as outlined below.

An objective of the learning algorithms is to determine whether some looks, or users who profile on particular looks will be more responsive to such discovery journey, and alternatively what other types of journeys.

Adjustment

Preferably, users can adjust look this into any of three key states, and intermediate points between key states. This represents up/down movement on vertical axis ("stylity"). Preferably, users can select, bookmark, or give strong negative to a look which feeds back to the algorithm.

The element of surprise in such adjustment is important to ensure that an unexpected look is returned by the engine occasionally (aka variable reward). To quantify "surprise", the look engine assumes that looks that are most distant from user's preferred look, will give the user the unexpected wow effect ("If nothing else works, why don't you try the following!"). Advantageously, the same algorithms as those described for discovery may be used (e.g. based on logarithmic spirals in look space), which is effective and saves time.

Networking

With reference to Figure 6, users A, B. etc. interacting with the look engine 820 application are provided with the option of uploading photos of garment items in their wardrobe which are then uploaded to a "virtual closet" 822 that represents a fashion repository. In essence, this is a global repository which can receive input items from multiple users and/or sellers. The virtual closet stores garment items and in this example it is connected to a seller 950. A garment item may include a photo, a description and owner identification details. The virtual closet may include new items (e.g. from a high street seller) as well as second hand items (e.g. from a user who scanned in items from the wardrobe).

Examples of use case scenarios ("Process Logic")

With reference to Figures 7, 8 and 9, the application groups user profiles into a number of categories, based on the profile criteria. In this example, categories include: A. Functional and Utilitarian Smart Buyers; B. Sunk Utility and Sunk Investment Buyers; C. Stuck-in-a- Rutter Buyers; D. Specific Searches Buyers; E. Rediscoverers (non buyers); F. Listeners and declutters (non buyers). Group A in this example is the group of users that want utility out of fashion and simplicity of buying, for example seeking either a tailored look or to buy for a specific use case/scenario), e.g. dress for a party on Friday or Monday meeting with a client. The application therefore, selects out of a catalogue of pre-defined customer journeys, a particular customer journey which corresponds to the needs of the customer segment as identified to the system by the user and shown in Figure 7: Scenario 2D ("Break it down"):

1) User Identification,

2) Itemisation of the look as supplied by the user into individual, buyable items,

3) Populating his/her basket with the said items, 4) Asking users to offload underused items via virtual closet component 822. With reference to Figure 7, Scenario 2A ("Use case") includes: 1) User Identification,

2) Presenting the users with the look options meeting the use-case,

3) Itemising said look into individual, buyable items,

4) Populating the users' basket with said items,

5) Asking users to offload underused items via virtual closet component 822.

Group B in this example is a group of users that wants to increase utility of idle clothing (sunk investment). These users are willing to scan in such "splurge item(s)" (that were expensive and are now mostly idle in their closet), add it to their Virtual Closet (the system's global database 822), which in turn feeds it to the look engine 820 to recommend (upsell) a bridge fashion item (product). This item, when bought will allow the user to wear the splurge item again because it matches up existing item(s) in the Virtual Closet with the user preference for (a) look/fashion and (b) current outside trends by the Look engine so that the listed item could be worn again. As shown in Figure 8, the process logic followed is such that: 1) User type is identified at Identification/Profiling,

2) Addition of the Splurge Item listing onto the Virtual Closet including a photo and an item description,

3) Look engine recommendation (up-selling) of the Bridge item. Group C in this example is a group of users that searches for a specific item. This is a group of users that know what they are looking for, be it an item of a look. As shown in Figure 9, the process logic is:

1) Users can override the Identification/Profiling stage by:

2) Providing input which may be specific, such as a social media pin, image, magazine photo, profile-survey based, and it is processed by the look engine 820. For example, the input can include picture of a look (which may be uploaded or selected by the user) or social media pin, 3) The application itemises the look into shoppable items as described above, before running the user profiling method,

4) The user fashion preferences are identified, and their personality is understood as Specific Searches by their behavioural signalling without the need for upfront profiling, which saves user's time,

5) An account is created for the user which may log in and a pre-filled questionnaire is offered to the user, and the user's style preferences are pre-populated by the Look engine. Price matching

With reference to Figure 10, the Look engine application includes a price matching module 823. The price matching module is tasked to call up items that match a user's requested look criteria (including criteria such as styles, colours, patterns) and personal criteria (e.g. shape, size) from all items available globally in Virtual closet repository. For example, two users may be considered: user A with low price sensitivity who asks for a high-street look vs a user B, highly price sensitive, also asking for the same look. The Look engine (together with the Look matrix) will be able to itemize different items for users A and B such that both will be offered the same look, however user B will be shown budget brand items, and/or second- hand items matching the look. User A, conversely, will see predominantly up-market and new items.

The "look engine" and "virtual closet" applications Figures 13 and 14 show screenshots of example applications according to aspects of the invention, as displayed on a display 801, which allows the user to interact with the system (e.g. via a touch screen for example). With reference to Figures 13A, 13B, a user logs into the application with user details imported from one of their social media accounts. The application then imports the user profile, with details such as name, gender, age, etc.

With reference to Figures 13C and 13D, the application prompts the user to input additional information in order to obtain a profile of the user's personality. With reference to Figures 13E to 13H, the user is prompted for further information on preferences regarding preferences in style ("look") and brands, which represent fashion criteria. With reference to Figures 131 and 13J, the application matches "looks" with the user profile criteria, as described above, and assigns a price (£150) to the "look" via the price matching module 823. The matched look includes a combination of second hand items (e.g. from the virtual closet repository) and high street items offered by sellers. The "looks" that match the profile criteria as displayed to the user. With reference to Figures 13K to 13P, the user is prompted to select a price range (by adjusting a slider in this example). In this example, the user increases the price range and the application selects a different look from the matched looks, which is within the selected price range (in this case at £350).

With reference to Figures 13Q, 13R, all of the matched looks may be displayed to the user, including "1990s vintage" and "trending look" in this example. The user is offered the option to view looks which are different to the originally matched looks by clicking the option "surprise me" shown in Figure 13S. Once this option is selected the application creates additional looks which may be a combination of an originally matched look with additional items from a seller. As shown in Figure 13T, in this example, the application presents to the user a "high street chic" look with a designer shoulder bag. With reference to Figures 13U to 13Z, the user has the option to select and replace individual items included in the look as well as the price range. Further, the user has the option to adjust the level of "risk" representing the appetite of the user for being presented with modified looks which are different to those originally matched to the user. In essence, the level of "risk" represents the level of matching tabulated in the matrix as described above.

The foregoing descriptions of embodiments of the present invention have been presented only for purposes of illustration and description. They are not intended to be exhaustive or to limit the present invention to the forms disclosed. Accordingly, modifications and variations will be apparent to practitioners skilled in the art.

Claims

1. A system for matching garment items, the system comprising a database comprising at least one user profile, wherein at least one user profile comprises two or more user profile parameters, the database further comprising a plurality of garment sets, each garment set comprising two or more garment items from said garment items, the system further comprising a processor for mapping a correspondence between the plurality of garment sets respectively to the at least one user profile.

2. A system according to claim 1, wherein the processor is configured to assign a first garment set to the at least one user profile.

3. A system according to claim 1 or claim 2, wherein for mapping a correspondence comprises assigning a garment set a location in a two or more- dimensional space corresponding to two or more user profile parameters.

4. A system according to claim 3, wherein the processor is further configured to calculate a distance between two garment sets located in the such space.

5. A system according to claim 4, wherein the space is Euclidian.

6. A system according to claim 4 or claim 5, wherein the processor is configured to map a curve in the two-dimensional space, the curve intersecting a predetermined garment set located in the space.

7. A system according to claim 6, wherein the curve is a logarithmic spiral.

8. A system according to any one of claims 6 or 7, wherein the processor is further configured to cluster the garment sets along the curve.

9. A system according to any one of the preceding claims, further comprising imaging and garment recognition means.

10. A system according to any one of the preceding claims, wherein the database comprises a plurality of images and/or parametric description for each item in the garment items to be matched.

11. A system according to any one claims 2 to 10, wherein the processor is further configured to map a correspondence between the first garment set and a further plurality of garment items.

12. A wardrobe organiser comprising a system according to anyone of the preceding claims.

13. A virtual shop assistant system comprising a system according to anyone of claims 1 to 11.

14. A method for matching garment items to a user profile, the method comprising the steps of: providing a plurality of garment items;

providing a plurality of garment sets, each garment set comprising two or more garment items from said plurality of garment items, and mapping a correspondence between the plurality of garment sets respectively to the at least one user profile.

15. A method according to claim 14, for use in a system according to any of claims 1 to 11.

16. A computer program product, loadable into a memory of an electronic communication device, and containing instructions which, when executed by the electronic communication device, cause it to be configured as the system of claims 1 to 11.