Shape Analysis Of Female Body Using Elliptical Fourier And Block Pattern Customization

The figure variation of human body is an important issue for flat pattern construction. Itis not only the foundation of mass garment production, but also the central task for developmade-to-measure garments. This thesis focused on using Elliptical Fourier Descriptors toanalysis the shape variation of the2D slices, Torso silhouette and3D form of female torso.A novel block pattern wireframe construction from3D body model and fatten algorithmwere also proposed. The correlationship between3D shape and2D pattern were discussedin detail. The paper innovation spot lies in the following aspects:(1) Feature extraction and measurementFirstly the format of3D body sample data and the preprocessing method weredescribed. Then the feature of human body was classified to six types. A set of commandtools which create or modify body features were defined based on computer geometrymethod. The proposed body represent model and algorithms were implemented as a COMplugin program for reverse engineering software named Rapdiform version2006. Thedeveloped software were design to extract customized features with customizable commandscripts, which can be applied to the human body measurement system for interactive orautomatic feature extraction. A new method for extract curve of the Minimum Girths ofupper body in bust plane was also introduced, which were importance feature curves forconstructing block pattern. The difference between Minimum Girths Upper Bust andMinimum Girths Upper Waist was compared using the sample set of this thesis. The frontview and side view of torso silhouette curves were derived by reslicing torso to horizontalslices.(2) Feature slices shape description, analysis and fittingThe horizontal slice curves of human body are closed curves and mostly similar to elliptical shape. Thus Elliptical Fourier is a good option for descript the slice shapeaccurately with lower less ellipse. Elliptical Fourier description of a curve is parameterized.Comparing to angle sampling discrete method it can be used to describe complicated curvesaccurately. Fitting error of bust slices were analyzed with different maximum harmonicnumber to establish the harmonic number used in this study. Principle ComponentAnalysis(PCA) and correlate analysis were employed to study the shape variation of slices.A new algorithm was proposed to construct symmetric slices. The first variation of slices iscorrelated with girth. The second variation is correlated with the width and thickness ofslice.The dimension of Elliptical Fourier Descriptors(EFD) was reduced with principlecomponent analysis. A regression model was built between principle components and inputparameters to generate slice curve from width and thickness parameters. The order of thepredicted girth errors of every slice from small to big is: waist girths, hip girth, bust girth.Because some component is correlated with slice girth but is not correlated with slice widthand thickness, the girth can’t be predicted precisely using only width and thicknessparameters. The predicted girth error is influenced by the input errors of width and thickness.The predict error is linear correlated to the input errors within±10mm. The predict errorwere1.2~1.8times input error of each parameter.(3) Torso silhouette and form variation analysisFirstly, the front view and side view silhouettes of torso were convert to closed curvesand were fitted using Elliptical Fourier. Principle analysis was used to show how eachprinciple component were related with shape variation. Torso form was resliced to a set ofhorizontal cross sections with equal distance, and each slice was fitted with EFD. Then thetorso’s shape variation was analyzed using PCA. Each principle component(PC) werediscussed in detail with graphical representation. The result shows that the first PC withlargest variance contribution of front view and side view silhouette are both correlated withbody weight. The ratio of should width and hip width has significant difference in frontview silhouette. Hip width changed from person to person even the bust width and waist width were same. The PCAresult of torso model shown the main modes of the torso’s shapevariation. The side view silhouette had more shape variation than the front view silhouette.The curvature of the vertebra varies greatly since it was changeable with body movementsand gestures. The fat distribution in the bust and abdomen area also varies among bodyshapes.A new horizontal sliced EFDs body model was proposed, and a new average shapecalculate method was apply using the model by calculate the mean value of EFDs and Zcoordinate of each slice. The verification was done in two aspects. Firstly, the meancoordinate of each landmark located on each sample was generated. The distance errors ofthe average landmarks to the surface of mean torso were analyzed. Secondly, thecircumstance of each corresponding slice of the mean torso was compare to the mean valueof the tested sample set. The circumstances of mean torso were smaller than the mean ofsamples. The mean torso of different samples was demonstrated, and the proposed methodcan be used to customize body form for garment manufacturer according to the positionedconsumer group. The flow of customization was discussed in detail.(4) Parameterization of3D torso and fitting from torso silhouettePCA were applied to the EFDs body model to reduce the dimension. Then multiplelinear regression analysis was done to parameterize the body model. Using the regressionmodel body model could be constructed with several body measurements, such as bust,waist, hip, etc. Since taking measurements usually need professional skills and is achallenge for customers. A new approach in this thesis utilized the front view and side viewsilhouettes shape to reconstruct the3D torso model. A regression model was constructedfrom the PCs of body model and the PCs of silhouettes. Samples were reconstructed usingthe regression model. The model was verified by comparing the scan model and thereconstructed model for the459female samples. the result shows that the reconstructedmodel were very similar to the scan model, especially the individual shape feature werekept. The torso length fitting error was very small and the maximum error was less than4mm. the average circumstance error is9.66mm and18.4mm in maximum. The reconstruct method could be apply to online made-to-measure system to generate3D models forcustomer from2D photograph captured with ordinary devices, such as a smart phone with acamera.(5) Gender recognition from torso silhouette using Elliptical Fourier DescriptorsThe silhouettes of torso have observable difference between male and female. A genderrecognition algorithm was developed in this thesis, which utilized the EFDs of silhouettescurves and artificial neural network. A BP neural network was constructed using theprinciple components of EFDs of silhouettes curves as input, and output is the genderclasses. The sample set was divided into train sets and test sets randomly. The inputsilhouettes were tested in three group, which are front view silhouette, side view silhouette,front and side view silhouette in combination. The experiment result shows that all groupsof input parameters have significant recognition accuracy, which were greater than98%.The recognition accuracy of the front view silhouette was the lowest, while the combinationof front and side view silhouette was the highest. The algorithm can be apply to manyrecognition areas which using torso silhouettes or images.(6)3D Basic block wireframe construction and block pattern flattenUsing the3D scanned body model, a3D basic block wireframe were defined andconstructed directly on the body model according to the clothing status of block pattern. The3D basic block was defined to be a convex hull of the body geometry. Then the wireframewas flatten to2D patterns according to a set of measurements. The flatten process wasdiscussed in detail and geometrically. The basic block of box type and tight waist type wereboth introduced. Virtual try on experiments were done using3D CAD software to verify thevalidity of flattened pattern. Bunka naked dummy and real scanned bodies were tested. Thedraping of flattened patterns fitted the model as expected and designed. The block patternflatten in this way had a similar form as currently widely used basic blocks. Such as newbunka basic block, Donghua basic block, etc. This increases the applicability to derivedgarment patterns easily from the block pattern. Another benefit is to develop individualblock patterns quickly and precisely, thus can be per potential used in made-to-measure systems.(7) Correlation analysis of3D body shape and2D block patternBasic blocks of459female samples were constructed and then flatten to block patterns.Statistics were analyzed and a regressive block pattern with bust girth was introduced. Therelationship of chest width of body and block pattern were described. New measurementswere suggested to be regressive parameters of bust dart other than bust girth, since thecorrelation coefficients were higher than bust girth. Measurements of block pattern wereanalyzed with3D body shape. Such as shoulder slops, front and back waist length, etc. Therelationships of body shape and block patterns were revealed. The conclusions of thisresearch had potential use in many research areas, such as block pattern customization,made-to-measure, pattern alteration, etc.