Automatic Morphological Classification with Case-Based Reasoning

Abstract

It is still a challenge for the apparel industry to develop good fitting products and underlying sizing and grading systems. This is due to the diversity of human bodies having the same traditional size but different morphotypes. Additional reasons are differences between different countries and special target groups such as young people or old people. The objective of the iMorph-approach is the morphological classification based on body scan data to be used for size system development and to provide better fitting clothes. Additional applications include recommendations systems in online business and curated shopping. iMorph is a unique approach to estimate the morphological classification of individuals based on body scan data. First, a morphological classification scheme was developed. It comprises 10 features and according ordinal scales. The analysis of the available body scan data showed that it is merely impossible to derive rules for the automatic morphological classification. Therefore, human experts visually classified a number of selected scans (data sets) by looking at the scanatars. A simple web-based application allows remote classification for invited experts. The resulting case base of classified scans is the core of a Case-Based Reasoning (CBR) system. It is able to compare the data of a new, unclassified scan with the scans of the case base. The most similar scans are the used derive a good estimation for the classification of the new scan. The crucial element of this iMorph approach is the applied similarity model. The functions are specific for each morphotype feature because not all measurements are relevant for all features. An example may illustrate this: for the classification of the shoulders of a new data set, only a number of scan data related to the shoulders is relevant, for example breast girth, distance shoulder to buttock. The classification is then derived from the morphological classification of these most similar "shoulder cases". The described approach has been proved to be valid and comprehensive [1]. It is flexible and extendable because each classification feature has separate similarity and retrieval functions with vast expert knowledge embodied and can be linked to various case-bases. Time consuming individual morphotype classification can be replaced by CBR technology and supports fashion product development as well as size recommendation in online retail.

Efficient virtual garment fit evaluation infrastructure based on synthetic avatar target customer groups for MtM application

“…The technology "3D-Side" is based on a principal component analysis (PCA) approach using all 3D scan material (>>10k scans) collected within the measurement series Size Germany [4]. It goes beyond the aim of this paper to present the working principles of PCA and its theory in detail, but it is important to understand that PCA acts in a certain way like a learning method, causing extreme data compression [5]. This means that AvatarStudio uses the most important aspects of the 3D shapes, and thus has "learned" the most important aspects of what human geometry looks like.…”

Landmarking and Measuring for Critical Body Shape Analysis Targeting Garment Fit

“…The apparel industry has directed much effort towards understanding body shape and while methods for effective classification of visual body shape are available [14,15,16,17], they are limited in that they do not directly affect pattern engineering or garment fit [18,19,20,21,12]. This is partly because the eye can be deceived with regard to shape, as illustrated in Figure 1.…”

A method for body posture classification of three-dimensional body models in the sagittal plane

“…Also, they are very precise in figure type analysis, but the posture classification includes only three possible types, Z, S, and I, which is very limited for made-to-measure clothing development. 21 Body posture is specified with bone size and structure and represented with the shape and place of the front and back body curve in the sagittal view. 20 While many existing algorithms have been focused only on detecting body posture irregularities, some algorithms are able to classify a given body posture into the one of several predefined categories.…”