Gpu-based tft-lcd mura defect detection method

A GPU-based TFT-LCD Mura defect detection method, comprising the steps of:

• (1) establishing a studentized residual based double-second-order regression diagnosis model based original image data, calculating first coefficients of the double-second-order regression diagnosis model, and fitting the first coefficients to obtain double-second-order regression background data;
• (2) obtaining influence quantities WK of respective data points on fitted values according to the original image data and the double-second-order regression background image data,
• wherein an influence quantity of an ith data point (zi, xiT) on the fitted values is
• WK
• i
• =
• z
• i
• -
• 
• (
• i
• )
• σ
• ^
• 
• (
• i
• )
• 
• p
• ii
• =
• (
• β
• ^
• 
• (
• i
• )
• -
• β
• ^
• )
• T
• 
• X
• T
• 
• X
• (
• β
• ^
• 
• (
• i
• )
• -
• β
• ^
• )
• σ
• ^
• 2
• 
• (
• i
• )
• ;
• a data deletion model of the ith data points (zi, xiT) is expressed as Z(i)=X(i)β+ε(i), where X(i) represents a data set obtained by removing xiT in an ith line from a coordinate data set X, Z(i) represents a data set obtained by removing zi in the ith line from a gray data set Z, zi represents a gray value of the ith data point, and xiT represents a coordinate vector of the ith data point;
• pii represents a diagonal element of a projection matrix P=X(XTX)−1XT generated by X;
• (3) excluding outliers and influential points in the original image data according to the influence quantities WK to obtain a new pixel point set ΦB;
• (4) establishing a double-N-order polynomial surface fitting model according to the new pixel point set ΦB, calculating second coefficients of the double-N-order polynomial surface fitting model, and estimating double-N-order background image data Z according to the second coefficients;
• (5) obtaining a residual image R=Z−Z+τ according to the double-N-order background image data and the original image data, and performing threshold segmentation on the residual image to obtain a threshold segmentation image;
• (6) performing morphological processing on the threshold segmentation image to obtain an eroded and dilated image,
• wherein the step (1) specifically comprises:
• (1.1) establishing a studentized residual based double-second-order regression diagnosis model Z=Xβ+ε=p00+p10x+p01y+p20x2+p11xy+p02y2+ε;
• (1.2) calculating estimation values {circumflex over (β)}=(XTX)−1XTZ of the first coefficients of the double-second-order regression diagnosis model by using a least square method;
• (1.3) fitting the estimation values of the first coefficients to obtain a double-second-order regression background image data =X{circumflex over (β)}=X((XTX)−1XT,
• wherein X represents a matrix representation of a horizontal and a vertical axis coordinates of a two-dimensional image;
• Z represents a pixel gray value at a corresponding coordinate position in X;
• p00, p01, . . . , p20 represent the first coefficients of the double-second-order regression diagnosis model;
• ε represents a residual value, which satisfies ε˜(0, σ2I), where I represents a unit matrix;
• β represents a vector representation of the first coefficients, and a confidence region of β at a level of 1−α is expressed as:
• ({circumflex over (β)}−β)XTX({circumflex over (β)}−β)≤p{circumflex over (σ)}2F(p, n−p, 1−α), where p represents a number of coefficients of the double-second-order regression diagnosis model;
• n represents a total number of image pixel points;
• {circumflex over (β)} and {circumflex over (σ)} respectively represent least square estimations of the first coefficients β and a variance σ;
• and F(p, n−p) represents an F distribution.