International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 05 | May 2024
p-ISSN: 2395-0072
www.irjet.net
Use of Machine Learning for Image Processing Nishita N. Merh1 1Student, Madhav Institute of Science and Technology, Gwalior, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This paper discusses about two approaches for
image processing. First is conventional digital image processing algorithm and second is the use of convolutional neural network (CNN) for finding image specific parameters. The proposed approach aims to automate the process of finding specific parameters of a given blob image having a normal distribution and using them for its classification. Key Words: Convolutional Neural Network (CNN), Digital Image Processing (DIP).
1. INTRODUCTION In the field of Artificial Intelligence (AI) and Machine Learning (ML), the “data” plays a vital role. This data needs to be processed, analyzed and inferences are drawn and learned. Then only any AI/ML based system can perform optimally. The data may belong to finance, engineering, science, or social fields. For analysis of any kind of data and for getting any inference out of it, a statistical approach is required. This leads to studying the distribution of data. Normal or Gaussian distribution is the fundamental distribution that is used across all the domains ranging from science, technology, medical to finance and social studies.
Fig -1: Background Grid Then a Gaussian image was generated using the equation of a 2D- normal distribution. The 2D Gaussian function used is given as follows [2]:
In this paper we will generate a Gaussian distribution of a blob and find it parameters viz. centre and spread, with conventional digital image processing. The same shall be compared with automated process of finding the parameters with the use of CNN. The results of the processing are presented.
Where, A is the amplitude, x0 and y0 is the center, and σx, σy are the x and y spreads of the blob. The image thus generated is as shown in Fig. 4(a).
2. METHODOLOGY
2.1 Conventional Digital Image Processing
The first step done was to generate the required synthetic images or blobs of the normal distribution. First a background image was generated having a 250X250 grid like structure shown in Fig.1.
The steps involved in curve fitting and determination of position and amplitude were: 1. Python library OpenCV (cv2) was used to load the grayscale image. The original image was loaded in grayscale format. [7][9] 2. Cropping was done using NumPy array slicing based on specified target dimensions. The image was cropped to a desired Region of Interest (ROI). [14] 3. Contrast Enhancement: Contrast Limited Adaptive Histogram Equalization (CLAHE) was applied. CLAHE enhanced the contrast of the cropped image, which further
© 2024, IRJET
|
Impact Factor value: 8.226
|
ISO 9001:2008 Certified Journal
|
Page 1025