• No results found

Journal of Environmental Science, Computer Science and Engineering & Technology

N/A
N/A
Protected

Academic year: 2022

Share "Journal of Environmental Science, Computer Science and Engineering & Technology"

Copied!
6
0
0

Loading.... (view fulltext now)

Full text

(1)

JECET; June – August-2013; Vol.2.No.3, 908-913.

Journal of Environmental Science, Computer Science and Engineering & Technology

An International Peer Review E-3 Journal of Sciences and Technology

Available online at www.jecet.org Computer Science

Research Article Research Article

JECET; June – August 2013; Vol.2.No.3, 908-913. 908

Role of Artificial Neural Networks in Digital Image Processing: A review

Sitendra Tamrakar and M. R. Aloney

Department of Computer Science & Engineering, Bhagwant University, Ajmer, (Raj) India Received: 19 July 2013; Revised: 16 August 2013; Accepted: 26 August 2013

Abstract: many studies have been done proving that artificial neural networks can perform significant role in processing of image. Here we will focus the application of artificial neural networks in image processing domain. We will discuss on how back propagation neural network (BPNN), Radial basis function neural network (RBFNN), General Regression Neural Network (GRNN), Probabilistic neural network (PNN), Complementary Neural Network (CMTNN) and Self-Organizing Map (SOM) are used for image processing for specific task. Key functions of artificial neural networks were discussed.

Keywords: ANN, Image Processing, steganography, GRNN, PNN, SOM, PSNR.

INTRODUCTION

Digital image processing is a part of study that uses a number of methods and algorithms to facilitate and understand the information that keeps a digital image.

Image processing roughly categorized into Low level and High level image processing: Low level image processing includes using of small information about contents of image. Image sharpening, image compression, edge extraction and noise filtering are coming under low level image processing. Where high level image processing tries to replicate human cognition and capability to create decisions based on information of the image content.

(2)

JECET; June – August 2013; Vol.2.No.3, 908-913. 909 In the last few years, artificial neural networks (ANNs) play an important role in digital image processing. The objective of this work is to wrap those methods get recognized and to demonstrate relationship between ANN techniques and digital image processing.

APPLICATIONS OF VARIOUS NEURAL NETWORKS IN IMAGE PROCESSING In this division, we will review some artificial neural networks that act as a tool for various image processing tasks like image compression, image Deblurring, Content-Based Image Retrieval, and noise classification in digital image

Image deploring using back propagation neural network: Image blur is challenging to stop in several predicaments and will typically destroy an image. Image deploring is the methodology of getting the definitive picture by utilizing the information of the debasing components. Debasement comes in numerous shapes, for example noise and blur. Fig. 1 showing bad effect of image blurring.

(a) Original Image (b) Blurred Image Fig.1: Effects of blurring on image

A major difficulty of existing restoration techniques for images is that they experience weak convergence attributes, the algorithm focalize to local minima, that they are unfeasible for legitimate imaging provisions. Added to its difficulty, a few techniques make prohibitive presumptions on the PSF or the correct image that restricts the portability of algorithm to diverse application. When using conventional technique, deblurring filters are connected on the corrupted images without the information of blur and its efficiency. Subashini et al.1 suggested an efficient method to overcome the problem of image blurring using Back Propagation Neural Network. In this work, theories of A.I. are directed for restoration difficulty where images are generally degraded by blur and corrupted simply by random noise. The suggested procedure embraced BPNN having gradient decent concept that includes about three layers.

This approach uses exceedingly nonlinear back propagation neuron pertaining to image recovery to have a good quality reconditioned image and also reaches quickly sensory working out, much less computational complexity as a result of much less variety of neurons utilized and also speedy convergence without extended training algorithm.

Image compression using back propagation neural network and self-organizing maps: Compression of digital image is a utility of digital data compression. Motive of compression an image is to reduce redundancy of data that contained by original image for storing and transmitting image in a capable structure.

Benefits of image compression include:

1. Potential cost savings in terms of carrier of image

2. It also reduces the chance of communication errors since fewer bits are transferred.

3. It also gives a stage of safety against illegal monitoring.

(3)

JECET; June – August 2013; Vol.2.No.3, 908-913. 910 Generally an image compression comes under following two categories:

Lossless (Reversible): In this technique, the basic image can be completely improved form the compressed image. These are also called noiseless since they do not add noise to the signal (image).

Lossless compression is used merely for some applications with severe supplies such as medical imaging.

Following methods are included in lossless compression:

1. Huffman encoding 2. Run length encoding 3. Area coding

4. LZW coding

General files types comprises GIF, TIFF, PSD and PNG.

Lossy (Irreversible): This schemes offer much superior compression ratios than lossless schemes. Lossy schemes are extensively used as the class of the reconstructed images is sufficient for the majority applications. Through this scheme, the decompressed image is not the same to the basic image, but logically secure to it2. This method rejects (loses) some amount of data and provide reduced size pictures.

General files types comprise BMP and JPG.

Panda et al.3 proposed a new algorithm using back propagation neural network (BPNN) to compress image. Their method includes taking a raw image and gives a compressed image as output.

Author show the application of back propagation neural network in image compression. The BPNN used for training and testing purpose for the analysis of various images. It observed from implementation result that the convergence time for the training of BPNN is much faster. Here several properties of compression like PSNR, compression ratio, bits per pixel are computed.

Amerijckx et al.4 find a way to compress a image using self-organizing Kohonen map. Their work focused on compression of original image by means of self-organizing map. The self-organizing Kohonen map is a consistent and capable way to accomplish vector quantization. They have proposed an innovative lossy compression method using of the organization property of Kohonen maps. Work based on the fact that successive blocks in an image are repeatedly similar, and thus coded by same code words through a vector quantization algorithm. They compared their work to JPEG type, and demonstrate their lossy compression scheme’s performance (in terms of PSNR) is better.

Image steganalysis using radial basis function neural network: Steganography and cryptography are two most famous techniques for hiding secret data within any media. Steganography is frequently compared with cryptography because it seems to be similar in terms their functionalities to protect secret data.

Basic difference between steganography and cryptography is the way to provide output. Output generated by steganography is not seems visible, on the other hand output processed by cryptography is jumbled so that it will be able to draw attention.

We can say that steganlysis is a procedure to identify existence of steganography5.

Sambasiva Rao Baragada et al.6 proposed a mixed technique of identifying steganography in a carrier image by fisher’s linear discriminant (FLD) function followed by the radial basis function (RBF) network.

(4)

JECET; June – August 2013; Vol.2.No.3, 908-913. 911 They have implemented steganalysis using FLD, RBF and mixture of FLD and RBF algorithms. They presented algorithm’s output for one steganographed image. Secret data is receiving by the projected algorithms with a range of degrees of accuracies. Their mixed method called FLDRBF is much capable in detecting the presence of secret information.

Noise classification in digital image processing using probabilistic neural network: Noise in image is an unwanted, random signal that goes in the communications system through the communicating medium and interferes with the transmitted message. Classification of noise is one of the major research areas of digital image processing. It plays an important role in science & medical fields. Noise classification in DIP is necessity because of choosing appropriate filter for smoothing and sharpening an image. Currently neural network identifies as a better substitute to conventional classifiers for several realistic classification problems. Santhanam et al.7 uses Probabilistic Neural Network to classify the noise present in an image after extracting the statistical features. For database they used ‘CASIA-Irisv3’ to check the performance of the PNN. Their experiments have been carried out in MATLAB. Proving that PNN is a better solution to classifying the noises than the MLP and BPN models.

Content-based image retrieval using self-organizing maps: CBIR is a method which uses visual contents to look for images as of large scale image databases in accordance with interests of users.

Content-based image retrieval also recognized as query by image content (QBIC) and content-based visual information retrieval (CBVIR) is the application of computer vision method to the image retrieval problem.

Fig.2: Diagram for CBIR system

Above figure8 demonstrate the diagram of typical CBIR system. The intent is to get discriminants which communicate as well as possible with the human decision for image resemblance and can be use in carrying out image queries. In CBIR, the images are indexed by characteristics straight derived from the visual content of the image. These characteristics usually comprise rather low-level data such as the textures, colors, and shapes9.

Jorma Laaksonen et al.9 introduced a new method for content-based image retrieval and a beginning assessment of its performance. They named PicSOM technique, in this method a Tree Structured SOM (TS SOM) is trained for every feature vector kind in use. The system then adapts to the user’s preferences by returning his further images from those SOMs where his responses mapped in majority closely.

Authors proved their work by results of experiments that the PicSOM is capable to efficiently choose as of a set of parallel TS-SOMs a mixture which moreover almost matches the greatest individual image map otherwise somewhat outperforms it in performance.

(5)

JECET; June – August 2013; Vol.2.No.3, 908-913. 912 Denoising of image using feed forward neural network: Image noise means unwanted signals that corrupt a digital image. That kinds of corrupted images needs processing of noise removal before it can be used for any program. Removal of noise from original image is called image demonizing. Demonizing of Image is still a difficult problem in the field of image processing. Demonising of image involves the treatment of the image data to create a high quality image. Image denoising is a significant job prior to further image processing, like feature extraction, segmentation, texture investigation etc. The principle of denoising is to eliminate the noise while retaining the edges and additional full features as much as possible.

Various methods are experimented to remove noise from image to enhance them. Some of the techniques are directly applied on the image data and some use the transformations like wavelet or frequency10. In this paper, we study use of feed forward neural network to demonising of natural image.

Sultan Uddin Ahmed et al.11 introduced a novel Image demonizing method by using feed forward Neural Network. They used prediction and generalization ability of neural network to find better quality image from noisy image. Local neighbour from 2×2 pixels and common value of them are measured for generating the input and aimed values of the training patterns to a neural network.

To get an improved image from the trained neural network, every pixel of the noisy image is recently generated by means of its four neighbours. Their method is verified with several standard images with diverse levels of noise. They examine that proposed approach is able to get better the quality of a noisy image in terms of Peak Signal to Noise Ratio. The visual examination suggests that the demonised image is appreciably fine than the noisy image.

DISCUSSION AND CONCLUSION

We have structured our study according to the functionality of neural network. A variety of survey has been done in this paper. We have discussed different image processing problems and its correspondence neural network. We focused our work on finding applications of various kind of neural network for image processing tasks like image deblurring, image compression, image steganalysis, image noise classification, content-based image retrieval and image demonizing. Neural network has done tremendous job in image demonizing field. Various kind of neural networks identify for several kind of noise removal. Our future work will focus on removal of image noise by use of artificial neural network.

REFERENCES

1. Dr.P.Subashini, Ms.M.Krishnaveni, Mr. Vijay Singh, “Image Deblurring Using Back Propagation Neural Network” WCSIT, 2011, 1, 6, 277-282,

2. Sonal and Dinesh Kumar, “A study of various Image Compression Techniques”, Guru Jhmbheswar university of science and technology, Hisar.

3. S .S. Panda, M.S.R.S Prasad, MNM Prasad, Ch. SKVR Naidu, “Image Compression Using Back Propagation Neural Network”, IJESAT, Volume - 2, Issue - 1, 74 – 78

4. C. Amerijckx, J.D. Legat and M. Verleysen, “Image Compression Using Self-Organizing Maps”, Systems Analysis Modelling Simulation, Taylor & Francis,2003, 43, 11,1529–1543.

5. Soumyendu Das, Subhendu Das, Bijoy Bandyopadhyay and Sugata Sanyal, "Steganography and Steganalysis: Different Approaches", IJCITAE, 2008, 2, 1, 200.

6. Sambasiva Rao Baragada, S. Ramakrishna, M. S. Rao, S. Purushothaman, “Implementation of Radial Basis Function Neural Network for Image Steganalysis”, International Journal of Computer Science and Security, Volume (2): Issue (1)

(6)

JECET; June – August 2013; Vol.2.No.3, 908-913. 913 7. T. Santhanam, S. Radhika, “Probabilistic Neural Network – A Better Solution for Noise

Classification” JATIT,2011, 27, 1,

8. http://www.cse.iitd.ernet.in/~pkalra/siv864/Projects/ch01_Long_v40-proof.pdf

9. Jorma Laaksonen, Markus Koskela, and Erkki Oja, “PicSOM: Self-Organizing Maps for Content-Based Image Retrieval”, Proceedings of IJCNN’99, Washington, DC, July 1999.

10. Sitendra Tamrakar, Dr. M. R. Aloney, “An evaluation of a few Image noise removal techniques”, AIJCSE, Volume I Issue I, Dec 2012

11. S.U. Ahmed, M. Shahjahan, M.M. Kabir, and K. Murase, "An Image Denoising Technique using Feedforward Neural Network", In proceedings of iFAN, Tokyo, Japan, September 26-27, 2010

*Corresponding Author: Sitendra Tamrakar; Ph.D. Scholar Computer Science &

Engineering Bhagwant University, Ajmer, India.

References

Related documents

These have produced hierarchical structures under various names to describe sources of risk, or risk categories or types, though these are usually focused on a particular project

We sought to identify blood-based, extracellular microRNAs 15 (ex-miRNAs) derived from extracellular vesicles associated with major stroke subtypes using clinical samples from

Home health aides typically work for certified home health or hospice agencies that receive government funding and therefore must comply with extensive regulations.. This means

We refer to a risk as process-related if its root cause is any combination of process behaviour (notably the activities performed and their sequence), resource behaviour (e.g.,

We use an aggressive matching strategy to find correspondences between an input glyph and a previously hinted template, considering both global and local features

Table 4 showed significant differences in the values of the plant yield due to the triangular interference between the irrigation methods, irrigation interval and

3 Section 3022 of the Patient Protection and Affordable Care Act (P.L. 111-148, PPACA), as amended, 4 directs the Secretary of Health and Human Services (the “Secretary”)

Finally, we realized that in targeting our own creativity, we were also able to foster a greater creativ- ity in our students as well as an increased sense of ownership and