Adaptive Thresholding Based Image Binarization Using Vhdl

The aim of this study is to recognize the given image with the existing image based on the technique of image binarization by MATLAB tool and it is simulated using VHDL (Very High Speed IC Hardware Descriptive Language) using MODELSIM tool. This image binarization is based on LEGION (Locally Excitatory Globally inhibitory Oscillatory Network) Concept. In this study the algorithm finds an optimum threshold technique, the other by separating the image background and foreground pixels. This algorithm has superior performance in separating the images from background in comparison with the other threshold techniques.


INTRODUCTION
To extract the required image from a given image, it may be separated into different components say foreground and background. An adaptive threshold concerning extraction of image from the background in a given image sequences is possible for implementation in hardware. The conventional histogram based threshold methods are deficient in detecting images due to poor contrasts between image and the background, or the change of illuminations (Otsu, 1979;Rosenfeld and De La Toree, 1983;Sezan, 1985;Trier and Taxt, 1995). But in real time applications, except adaptive thereshold techniques are not fast enough for hardware implementation (Kittler and Illingworth, 1986). The image binarization (Lee et al., 1991;White and Rohrer, 1983) is a process of dividing the original image into two components, foreground or image and background. The gray level value of background (value '0') is different from the foreground (value '255') values. So, for the effective separation of foreground and background, this thresholding techniques may be considered (Hertz and Schafer, 1988;Sezign and Sankar, 2004). The theresholding techniques yields a binary image as an output by assigning pixels with values less than the threshold as 0's and remaining pixels as 1's. In this study the Image binarization is carried out by MATLAB tool and it is simulated using VHDL (Very High Speed IC Hardware Descriptive Language) using MODELSIM tool (XESS Corporation, 2001). This image binarization is based on LEGION (Locally Excitatory Globally inhibitory Oscillatory Network) Concept (Ligon et al., 1998). The LEGION concept has the following features: • Binarization is realized in a very short time due to the simple processing. • Due to the simple structure of each cell representing a pixel compact integration of many cells on a single chip becomes possible. • Fast software implementation is possible.

Neural oscillator concept:
There are three possible ways to reach synchrony as shown in Fig. 1 to 3.
LEGION is composed of a Basic Oscillator, Local Excitatory Connections (produce phase synchrony within each object) and Global Inhibitor (it receives inputs from the entire network and feeds back with . Each Oscillator will respond to a detected feature at some location in the image. Synchronized group of oscillators is separated from other synchronized groups that represent different objects by de-synchronization. It is proposed in Digital implementation, Model described by digital algorithm, both gray/color scale binarization and segmentation in a single algorithm (Yanowitz and Bruckstein, 1989). Reprogrammability is possible using FPGA (Shirazi et al., 1995).
The Fig. 4 shows Digital LEGION Image Binarization Architecture. It consists of six blocks namely:

Identification of dependent cell (or) local excitation:
If pixel Ii is a neighbor of X i = 1 and WI i Pi>T2 then Ii is a dependent cell of X i .
Where, WI i P i = weight between pixel I i and weight corresponding to X i = 1 location.

RESULTS AND DISCUSSION
The following ( Fig. 6 and 7) are the sample image and its binarization.
The sample image is given to MATLAB Tool and the Corresponding Matrix values are obtained. The following tables (Table 1 to 6) shows the output of the (Input ROM, weight calculation, leader cell 1 and 2, local excite and output of RAM respectively) in the digital LEGION image binarization architecture. The weight value between 36-24 is 19, 127-36 is 2, 127-24 is 2 and 127-128is 127. Let the leader threshold value be 1000.
The corresponding simulated output using VHDL is shown in Fig. 8.

CONCLUSION
In the present study, the image binarization is done by MATLAB and the same is simulated using VHDL by using MODELSIM. In this study the algorithm finds an optimum threshold technique, the other by separating the image background and foreground pixels. This algorithm has superior performance in separating the images from background in comparison with the other threshold techniques.