An improved image processing algorithm for automatic defect inspection in TFT-LCD TCON

3.1 Medical image enhancement

In the process of medical image generation and transmission, inevitably affected by the external environment and generate some noise, affects the final medical image effect, so to guarantee the effect of the medical image, the effects of reducing noise, medical image enhancement algorithm arises at the historic moment, medical image enhancement algorithm, not only can effectively reduce the influence of the noise, and can also enhance medical image clarity, achieve the visual effect. On medical image enhancement algorithms, there are many different ways, some of these methods in medical image denoising, some in gray medical image enhancement and color medical image enhancement has prominent effect in different aspects, such as, according to the relevant data contrast found Jourlin and Pinoli study algorithm is more general, on the one hand, can effectively abate the noise, on the other hand, can improve medical image clarity, sound processing medical image detail, and also can enhance the medical image grayscale and color, at the same time the medical image is not so and distortion, is now a better calculation method in the field of medical image processing [17].

3.2 Edge detection

Medical image processing, edge of mishandling would happen often lead to a medical image, medical image edge is due to the uneven distribution of medical image grayscale, texture or caused by improper handling, at this time to detect the edges of the medical image, the edge method also has many, but this among them, the Canny method is the most ideal effect, but this kind of algorithm for noise treatment effect is bad, therefore, the improved Robert comes edge detection algorithm, the improved Robert algorithm overcomes the problem of improper handling Canny method for noise. At the same time, it overcomes the disadvantage of the traditional Robert algorithm in rough medical image edge processing and can make the processed medical image have a clear viewing effect.

3.3 TFT-LCD system structure

TCON timing signal is generated based on the needs of the TFT-LCD Panel. To understand the principle of TCON, one should first understand the display principle of the LCD Panel. The internal structure of a typical TFT-LCD panel is shown in Figure 2. Liquid crystal has light transmittance controllability, change the voltage applied on both ends of liquid crystal; liquid crystal transmittance will change accordingly.

Figure 2

Contrast adjustment relationship

A liquid crystal unit displays a sample point, so a medical image can be displayed if it is arranged in a matrix according to the data structure of the binary medical image into a crystal array. In general, LCD arrays can be arranged in two ways: Strip and Delta. Different arrangements determine different Red Green Blue (RGB) sampling orders, and a pixel consists of three liquid crystal units (RGB). Therefore, the number of liquid crystal elements in the liquid crystal array determines the display resolution. TFT-LCD is formed by separating display pixels and scanning electrodes through TFT transistor on the basis of super-distorted STN, which is characterized by overcoming the crossover effect of STN-LC9 [18].

TFT-LCD system consists of two parts: LCD control module TCON and LCD panel module. The LCD display adopts the active matrix driving mode according to the row and column. The row lines are connected to the gate of the pixel NMOS tube, and the column lines are connected to the Source of the NMOS tube. In the LCD module, the line and column line are driven separately. The circuit that drives the column line is called the Source Driver. The source driver and the gate driver together control the charging and discharging process of the liquid crystal unit. When the scan signal is valid, all TFT cells on this line are turned on at the same time, and the RGB voltage charges the memory capacitor through TFT. Therefore, the voltage difference between RGB and Vcom determines the voltage of the liquid crystal column. When the scanning signal is invalid, the TFT unit on this line is disconnected, and the charge of the storage capacitor can remain basically unchanged in one frame time, thus achieving the static display effect of the duty cycle approaching 100%. The main function of TCON is to provide the necessary timing control signals for the source and gate drivers in TFT-LCD panels. According to the requirements of the Panel, the most basic output signals of TCON are STV, OEV, CPV, STH, CPH, OEH, PLC, etc. In addition, in order to achieve different display modes, there are L/R, U/D, MOD, Q2H, and other control signals [19].

3.4 TCON module design of timing controller

TCON module of timing controller includes PLL, I2C interface, DIV of the frequency divider, HSB of line synchronization module, VSB of field synchronization module, and CSB of control signal output module. The field synchronization module VSB generates the control signal needed by the gate driver. Line synchronization module HSB generates the control signal required by the source driver. The control signal output module CSB generates the scanning direction control signal.I2C interface is introduced in the design of this system. The control signal can be controlled through the internal register, and the change of display mode can be realized by changing the software, thus greatly reducing the number of pins of the external hardware circuit. PLL can provide system clocks with 1920, 1440, 1200, and 1152 resolutions, while 960, the corresponding clocks of 480, 400, and 320 resolutions are obtained by frequency division of the DIV module [20].

Since output control signals required by different LCD resolutions and working modes are different, TCON should first make a judgment and choose the correct working mode to generate appropriate control signals. These options include LCD resolution, external/PLL clock mode, split/composite mode, NTSC/PAL mode. LCD resolution selection: in the actual display system, TFT-LCD has different sizes and resolutions. The resolution of general medium and small size TFT-LCD is 480×234 (2.5″, 3.5″), 960×234 (3.6″, 5″, 6.4″), 1200×234 (6.5″), 1400×234 (6.2″, 7″), 1920×234 (9″), etc. For different resolutions of LCD, the need for some control signals will be different. External /PLL clock mode selection: the working clock source is provided in two ways, PLL mode, and external clock mode. In the PLL working mode, the VCO circuit generates oscillations, which are transferred to TCON by a phase-locked loop circuit a working clock for stability; in external working mode, the working clock is provided externally.

Separate/compound mode selection: the input synchronization signal can be divided into compound synchronization signal and separate synchronization signal. The synchronization pulse flag bit of these two kinds of synchronization signals is different. NTSC/PAL options: NTSC and PAL options have completely different lines per frame and sync pulse flags, so a decision should be made before the input signal is processed and the control signal is generated. Display mode selection under high resolution: there are 8 different display modes under high resolution, including FULL, Center, Wide, Zoom1, Zoom2, Left, Right, and Zoom3. Under different display modes, the display area, output control performance, and pulse width will be different.

4.1 Contrast adjustment

The contrast of the display is defined as: in a dark room, the brightness under the white screen is divided by the brightness under the black screen so the brighter the white and the darker the black, the higher the ratio. Contrast enhancement adjustment tries to increase the contrast between different colors, making bright pixels brighter and dark pixels darker. Contrast reduction reduces the contrast between colors, darkening bright pixels and brightening dark ones. There are two ways to change the contrast in an LCD display:

1. The contrast of an LCD panel. High contrast means a relatively high level of brightness and color. In fact, it is determined by the optical properties of LCD back-light and LCD panel. Once determined, it cannot be adjusted.

2. The contrast of the image itself reflects the distribution of the image from light color to dark color, and reflects the transition characteristics of the image in the gray level. This article discusses the contrast of the image itself. There are two methods of image contrast enhancement: spatial domain method and frequency domain method. After dividing the image into target area, background area and transition area through min value segmentation, the piecewise linear transformation method based on double min value segmentation can adjust the image contrast by adjusting the weight value W. The method is not only simple to adjust, but also considers the information characteristics of the image itself.

In the implementation of the circuit, since every frame of image in the continuous video stream always enters the contrast processing module one pixel at a time according to the order of horizontal and vertical scanning, to calculate the current frame's min value, it is necessary to store one frame of image. However, due to the continuity of the image, there is no need or need to increase the frame memory. We can use the value of the previous frame to adjust the contrast value of the current frame, which will not affect the adjustment effect.

Contrast adjustment refers to Figure 2. Contrast adjustment mode is shown in Figure 2. Here mainly involves the design of a multiplier, in view of the requirement of real-time, we adopt a pulsating array, due to the combination of the multiplier delay is bigger, can’t meet the requirements of our high speed, so we skillfully to pulse array of joined the level 4 trigger to form the structure of the assembly line, to reduce the delay time of each level. The L1, L2 and L3 in the figure are limited in scope, L1L3 ∈ {1/42/43/415/46/47/424}, the precision of L2 is relatively high, because it involves the connection problem of curves where L1 and L3 are located. We use 8 bits, L2 ∈ {0.0000000 ~ 1.1111111}, namely {0~1.9921875}.

The corresponding functional expressions are as follows: (1) to (3)

4.2 Gamma correction

4.3 To shake

In many practical applications of LCD displays, affected by the accuracy of specific display terminals, the digital images represented by multi-bit bits in the processing process need to reduce the number of pixels before display to meet the requirements of display terminals, which results in a problem of conversion from high-precision display space to low-precision display space. This paper firstly analyzes the shortcomings of the basic methods to realize this transformation - direct interception. In order to obtain higher image display quality, the commonly used error diffusion halftone algorithm is discussed.

The appearance of "block effect" is intolerable to the display system because it greatly reduces the display effect of the image. The research on the principle of human eye vision shows that the ability of human retina to recognize the gray level is 1, that is, under the condition of 100 gray levels, if there is a change of 1 level, it can be recognized. Therefore, when an 8bits of gray scale information is converted to 6bits, its gray scale change rate is 1/64(1.56%). When an 8bits of gray scale information is converted to Obits, its gray scale change rate is 1/16 (6.25%). Therefore, the algorithm directly intercepting the redundant bits will inevitably cause the image quality recognized by human eyes to decline. However, in terms of the cost of hardware implementation, the 6bit LCD panel can reduce the cost compared with the 8bit LCD panel, and the pseudo-true color panel can reduce the maximum number of driving IC, thus reducing the difficulty in visual Angle and contrast design. Therefore, in many applications, such as small and medium sized LCDS equipped with portable devices and mid - and low-end desktop LCDS, fake color panels with 6 bits or less are used. As a result, there is a contradiction between saving cost and greatly reducing image quality.

In order to improve the display quality of the image as much as possible, many algorithms are proposed, which are uniformly classified as digital halftone technology. The basic function of halftone processing is to eliminate the stepped striations caused by the discontinuous tone output of the image, so as to achieve the effect of image color enhancement, and effectively eliminate the inherent block-effect defects of the image, so as to make the image smoother and softer, and further improve the quality of the image. The basis of digital halftone processing is that the human eye has a "spatial proximity characteristic" at the initial stage of processing visual information, so that spatially adjacent parts of the image are treated as a whole. In this way, the human eye has a low resolution of details within a certain visual Angle (also known as the low-pass property), and at the same time, the perception of the gray or color of a small facet by the human eye is the perception of the total gray or color of the facet at a certain distance. The gray image or color image is binaries by using this characteristic of human eye. As long as the binary pixel points are close enough in space, the human eye can see the continuous tone.

In halftone processing, gray is not provided by the halftone all source image gray level, there is a similar problem, with an approximate gray instead of the original gray level, there must be error, and simply give up, if this error is bound to affect the image processing effect, such as the change of the grey value and color value of jumping, etc. Therefore, how to deal with the error becomes the key of the problem, and the error diffusion algorithm comes into being.

Error diffusion is based on a single principle. Once a pixel is quantized, it has errors that affect the pixels around it. The process by which the error affects the surrounding pixels is called diffusion, which means that the error is divided into components that are added to the gray scale values of the adjacent pixels. A negative feedback system is the system in which the system is in a self-correcting state through the diffusion of errors. This algorithm is a neighborhood processing process, which diffuses the quantization error of the current pixel to the neighboring pixel in a certain proportion. In this way, the local quantization error is a compensation for the adjacent pixel points, so the error diffusion system has the ability of self-correction. Moreover, it can guarantee that the total gray scale or total color of the image before and after processing is consistent on the whole.

Simple dithering can be achieved by using dithering matrix, like a typical BAYER matrix. However, the medical image processing effect of this scheme is not very good. After matrix processing, the medical image produces block effect. Therefore, we adopt the better Floyd-Steinberg error diffusion algorithm. We adopt {1/16, 5/16, 3/16, 7/16} coefficient combination Floyd-Steinberg error diffusion algorithm based on delta-sigma modulation. Noise-shaping feedback coder is a DSM for word length compression, and f–s algorithm is one of NSFC. The algorithm of NSFC is shown in Figure 3.

Figure 3

Algorithm structure diagram of NSFC

The corresponding transfer function is: Y(x) = X(z) + [1 − H(z)]* N(z)

The noise transfer function corresponding to Floyd-Steinberg algorithm is: NTFz1, z2) = 1 − 1/16(z1⁻¹z2⁻¹ + 5z2⁻¹ + 3z1z2⁻¹ + 7z1⁻¹)

Error diffusion algorithm is mainly composed of column and column counting module, control module, data path and First Input and First Output (FIFO) memory. The control module solves the sequence of reading and writing of FIFO. The data path mainly completes a weighted average of error diffusion coefficient to obtain the current pixel value and calculate the corresponding diffusion error. FIFO is used for error storage of adjacent two lines. Given that the error can be negative, a 3bit FIFO is used to store the signed error. Dual-ports Random Access Memory (RAM) automatically generated by VeriSilicon is adopted to produce FIFO that can read and write simultaneously.

4.3.1 Simulation results and discussion

In order to verify the effectiveness of the above algorithms, we respectively used MATLAB language and Verilog language to simulate some algorithms of contrast adjustment, brightness adjustment and dithering, and provided the core code of MATLAB simulation and the comparison before and after medical image processing.

Figure 4, presents the spectral response of the images. An image offers resources as a global shutter and control of exposure time. Either the start or the exposure time can be controlled by parameters transmitted through interface. There are also parameters available to adjust single image acquisition or acquisition of continuous images. The start of exposition can also be controlled through a signal generated externally, and that is applied to entrance. The proposed algorithm is described below and simulation results are explained.

Figure 4

Spectral response of RGB extraction

Figure 5 represents the experimental analysis of proposed scheme to evaluate the sparsity rank under various values of ρ for different defects in images. On the other hand the evaluated running time for various defects in images is depicted in Figure 6.

Figure 5

Analysis of sparsity rank under various values of ρ for different defects in images

Figure 6

Analysis of running time under various values of ρ for different defects in images

The evaluation of running time is an important parameter that is essential for the manufacturing process of industrial application. Figure 5, depicts the relation between the sparsity rank and values of ρ. Here ρ is considered as {2, 4, 8, 10,. .., 78, 80} and the corresponding sparsity rank is calculated as SR_k, where k = 1,2,3,. .. ,length (ρ). Length (ρ) represents the length of vector ρ.

The experimental analysis of designed model, which is implemented in FPGA reveals that with much improved modules, it can be substituted in commercial LCD. The practical implementation of proposed model reveals that it can be substituted with LCD for obtaining video image at high speed and absolute improved definition. MATLAB simulation results are as follows: "LCD output image without correction" is smaller than the gray value of the source image, and the image appears darker, while the image processed by the algorithm proposed in this paper is darker. The "corrected image" has a higher gray value than the source image, and the image appears brighter, while the image processed by the algorithm proposed in this paper is brighter. The "corrected LCD output image" processed by the pre-modified sub-channel correction method proposed in this paper is closer to the original image than the "corrected LCD output image" processed by the Gamma=2.2 correction method. Compared with the 8bit image, the 6bit image is slightly fuzzy, and there is a lack of detail in the image, but basically it has achieved a better effect. The processing effect of Floyd-Steinberg diffusion matrix method is better than that of Bayer jitter method, and the "block effect" is further improved. The dynamic error diffusion method based on the edge has the best processing effect, and its "block effect" is controlled to the maximum extent. Moreover, the edge details of the image are richer and finer than those of other algorithms, and the image edge is clearer.