Sp100402

Post on 14-Jun-2015

264 views 0 download

Tags:

description

Identification of Animal Fibers with Wavelet Texture Analysis

transcript

Centre for Material and Fibre InnovationCentre for Material and Fibre Innovation

Identification of Animal Fibers with Wavelet Texture Analysis

Dr Junmin ZhangDr Stuart PalmerProfessor Xungai WangCentre for Material and Fibre InnovationDeakin UniversityAustralia

Centre for Material and Fibre Innovation

Introduction - Cashmere

As cashmere processing capacity outstrips available supplies of cashmere, some processors use superfine merino wool to blend with cashmere

Cashmere wool blends provide the high quality worsted suiting fabric and produces a lower cost product while exploiting the positive market perceptions associated with the luxury cashmere content

Source: http://en.wikipedia.org/wiki/File:Old_O102_cropped_small.jpg

Centre for Material and Fibre Innovation

Introduction - blends

Current standard test methods for analysing blends of specialty fibres with sheep’s wool are based on scanning electron microscopy

The current operator-based methods are tedious and subjective

It is desirable to develop an objective, automatic method to identify and subsequently classify animal fibres

Source: http://commons.wikimedia.org/wiki/File:Australian_Cashmere_Goats.jpg

Centre for Material and Fibre Innovation

Introduction – fibre classification

Centre for Material and Fibre Innovation

Introduction – fibre classification

Centre for Material and Fibre Innovation

Sample image preparation

CashmereMerino wool

Centre for Material and Fibre Innovation

Sample image preparation

CashmereMerino wool

Centre for Material and Fibre Innovation

The 2D dual-tree complex wavelet transform

Centre for Material and Fibre Innovation

Fibre surface feature extraction

Detail images represent thecontent of successive frequency bands

The final approximation imagecontains the low frequencyvariations in backgroundillumination

Centre for Material and Fibre Innovation

In this work, the analysis object is the fibre surface, and the texture feature is defined as:

Where M×N is the size of the fibre surface image, and are the pixel grey-scale values of fibre surface image in scale s and direction k.

Fibre surface feature extraction

M

i

N

j

ksNMjk kjiDE

1 1

21 75,45,15,

jiD ks ,

Centre for Material and Fibre Innovation

Fibre surface feature extraction

From the Cashmere Fiber Distinction Atlas, 13 cashmere fibre images and 15 merino fibre images were prepared

From each of the 28 fibre images a texture feature vector consisting of 24 (6 orientations x 4 scales) energy features was developed

Principal component analysis was used to reduce the dimension of the texture feature vector

Centre for Material and Fibre Innovation

Principal components analysis

Centre for Material and Fibre Innovation

Discriminant analysis

Round 1 2 3 4 5Training set size 28 26 24 22 20Training correct no. 27 25 23 22 20Testing set size 0 2 4 6 8Testing correct no. 0 2 4 5 6

Centre for Material and Fibre Innovation

Allied work – automatic pilling classification

Centre for Material and Fibre Innovation

Future work

1. non-linear neural network classification;

2. testing of the performance of the wavelet texture analysis method of fibre identification on a larger set of real cashmere and other fibre samples; and

3. the application of the wavelet texture analysis method to the of task analysing/assaying blends of specialty fibres

Centre for Material and Fibre Innovation

Thank you for your time