T. Anil Raju1*, K. Srihari Reddy1, Sk. Arifulla Rabbani1, G. Suresh1, K. Saikumar Reddy1 and K. Girija Sravani2†
1 Department of Electronics and Communication Engineering, Lakireddy Bali Reddy College of Engineering, Mylavaram, Krishna District, Andhra Pradesh, India
2 Department of Electronics and Communication Engineering, KL University, Green Fields, Andhra Pradesh, India
Abstract
Image processing continues to enable the technology revolution that we are experiencing today. The main objective is to carry out a new cooperative approach for image segmentation and feature extraction using DSP Processor with good accuracy. The basic process involved in this paper to understand image segmentation is Edge detection. To improve the segmentation process and better estimation of the final clusters centers, involvement of DSP processor is used for computation and accuracy. Image segmentation using DSP processor TMS320C6748 is implemented in this proposed work. After detecting the edges of the image, Segmentation is done based on Morphological operations. The performance is measured by including some various optimization techniques and all results are shown using CCS platform. Hence our proposed method shows that the segmentation computation and accuracy is improved.
Keywords: Image segmentation, DSP processor TMS320C6748, thresholding method
Image processing is an active research area which has become a vital component of a large number of present real-time applications. Segmentation is an important imaging technique for detecting abnormal changes in intensity levels of image. In computer vision, image processing is a highly challenging field; image segmentation is still a broad research topic and also an essential task in many applications because it plays a major role in the MRI image understanding. The MRI Images segmentation [1] has become the focus of contemporary research; some applications require a simple division of image into homogeneous regions while others require more accurate detection [2]. For this reason we need a special unit that we use for digital signal and image processing in a real-time execution and also the high-speed data transfer.
A digital signal processor [3] is one of the core technologies in rapidly growing application that become a key to many domains. Digital signal processors [4] such as the TMS320C6x family of Texas Instruments are like fast special-purpose microprocessors with a specialized type of architecture and an instruction set appropriate for signal processing. All the DSP methods are successful because of the combination of both hardware and software. The proposed algorithm will implement practically on our hardware.
MRI is a modern imaging technique that gives a more detailed image than CT scans and X-rays (Figure 12.1). MRI scans uses a large magnet field or radio waves to take images of the brain and other structures of the body. High spatial resolution and excellent soft tissue diagnosis are the advantages of MRI over other medical imaging techniques. In our research, the input MRI images are simulated with T1-weighted contrast and all images are obtained from the Brain web database from McConnell Brain Imaging Center of Montreal Neurological Institute at McGill University.
Segmentation is the division of an image into regions or categories which correspond to different objects or parts of objects [4]. Every pixel in an image is allocated to one of a number of these categories. It is an essential step in image processing and also it is a critical stage of image. Image segmentation methods can be classified into three categories [5]: Edge-based, region-based, and pixel-based methods. Clustering and edge detection are important tools for a variety of medical applications.
The processor which is going to be used is a TMS320C6748 DSP processor. This is a high-level processor which provides many features. It contains a 5v in power supply which will be given through an adaptor. It contains USB and OTG ports for connecting different devices. This board contains SATA (Serial Advances Technology Attachment) which is an interface for transferring the data between CPU and storage devices. A reset button is also present in this board for resetting the device.
An Ethernet port is also available for connecting to internet. This board provides some user buttons for the performing of different actions. The board is connected to a computer through JTAG. This board comes with a camera sensor through which we can capture images directly from the board. Along with camera sensor, a mic sensor is also present in this board through which we can take input speech signals directly. In accordance with these a fingerprint sensor is also available with this board which is only available in these versions. A VGA (video graphic array) port is also available through which we can connect the output to the display of the computer directly (Figure 12.2).
Code composer studio software is an integrated development environment that supports Texas Instruments microcontrollers and embedded processors. Code composer studio software combines the advantage of eclipse software framework with advanced embedded and debug capabilities. There are some better features available in the Code Composer Studio. They are Program Code Editing, Application Building, and Application Debugging Features. CCS allows ‘C’ and Assembly language program. The Application Building Feature of the CCS is the basic process which is used to create, build, debug, and test the programs. The debugging features of CCS are loading data, FILE I/O, watch window, graphing signals, setting up of break points and usage of probe point. These languages save valuable time in the design process. The C and C++ code can be compiled as well as built to generate the desired file. The project we create keeps track of all the information regarding the input image we provided before preprocessing.
Thresholds in those algorithms may be decided on manually in step with a priori understanding mechanically through photograph information [6–14]. Algorithms may be in addition divided to facet-primarily based totally ones, region-primarily based totally ones and hybrid ones. Thresholds with inside the facet-primarily based totally algorithms are associated with the facet information. Structures are depicted via way of means of facet points. Common facet detection algorithms such as Canny facet detector and Laplacian facet detector may be categorised to this type. Algorithms try to locate facet pixels even as they get rid of the noise influence.
For example, Canny facet detector uses the brink of gradient value to locate the ability facet pixels and suppresses them through the tactics of the non-maximal suppression and hysterics shareholding. As the operations of algorithms are primarily based totally on pixels, the detected edges consist of discrete pixels and consequently can be incomplete or discontinuous. Hence, it is very important to use post-processing like morphological operation to attach the breaks or get rid of the holes. The technique has the cap potential to phase 3-d photographs with true accuracy; however, the drawback of this technique is the issue of the technique to system the photos of textured blob objects. This can be represented by the block diagram shown in Figure 12.3.
The proposed system has been developed to solve the problem of initial parameters for the level set algorithm such as the initial contour and its centre. Thresholding is a superb way to extract beneficial records encoded into pixels at the same time as minimizing heritage noise. This is performed via making use of a remarks loop to optimize the edge fee earlier than changing the authentic grayscale picture to binary. The concept is to split the picture into parts, the heritage and foreground [6].
The simplest property that pixels in a region can share is intensity. So the natural way to segment those intensities is through thresholding, the separation of light and dark regions. Thresholding creates binary images from grey-level ones by turning all pixels below some threshold to zero and all pixels about that threshold to one. (What you want to do with pixels at the threshold does not matter, as long as you are consistent.) If g(x, y) is a thresholded version of f(x, y) at some global threshold T,
Morphology is nothing but the study of shape and structure of the image. Morphological image processing is a collection of non-linear operations related to the shape or morphology of features in an image. A morphological operation of a binary image creates a new binary image in which the pixel has a non-zero value only if the test is successful at that location in the input image.
In order to improve the total execution time on the chip, various optimization techniques were used to check the performance: C code, CCS compiler and cache memory. The TMS320C6748 DSK provides a feature to change the size of cache memory from 0 KB to 256 K Bytes; it can be made by changing the parameters in the General Extension Language (gel file). The number of clock cycles required for each function is calculated and compared by a method called profiling for benchmark. In our methodology of profiling, all of its functions were analyzed and tested to determine the values of clock cycles. The improved performances were after adding the presented optimization technique.
To implement the Image Segmentation using code composer studio, the following steps are to be followed:
In this section, some of the brain MRI images containing tumor taken for testing our proposed algorithm are shown below.
The results which are obtained by processing the image only by the software platform without involving hardware like DSP Processor is sensitive to noise. But results obtained by involving DSP Processor is not sensitive to noise and it improved the quality of classification (see Table 12.1).
Table 12.1 Performance improvement achieved.
n° | Using optimization techniques | ||
---|---|---|---|
Function | No. of CPU cycles (without optimization) | No. of CPU cycles (with optimization) | |
1 | Image conversion | 3924756 | 3131025 |
2 | Median filter | 91295273 | 73084236 |
3 | Dividing original image | 95877799 | 90436952 |
4 | Average means of new images | 3402892 | 2935339 |
5 | Calculating new threshold | 98642594 | 92527099 |
6 | Comparing new and old threshold values | 98623241 | 91348615 |
7 | Obtaining output image after processing | 93635675 | 89908543 |
This chapter proposed a powerful new approach for segmenting brain tumors and calculating their area using a Thresholding algorithm. The proposed method is tested on scanned MRI images of the human brain, and if there is noise, the non-uniformity is removed using a median filter and also by a modified new threshold value. The results show that the proposed approach to the detection of brain tumors may improve the quality of image segmentation and produce high diagnostic accuracy. Finally, you can make a rough deduction to calculate the shape of the tumor.