Behaviour Analysis using Handwritten Data

Page 1

GRD Journals- Global Research and Development Journal for Engineering | Volume 5 | Issue 3 | February 2020 ISSN: 2455-5703

Behaviour Analysis using Handwritten Data Premraj Joshi Department of Computer Engineering Dr. DY Patil Institute of Technology,Pimpri Simran Raut Department of Computer Engineering Dr. DY Patil Institute of Technology,Pimpri

Siddharth Kota Department of Computer Engineering Dr. DY Patil Institute of Technology,Pimpri

Aishwarya More Department of Computer Engineering Dr. DY Patil Institute of Technology,Pimpri

Prof. Pradip Shewale Department of Computer Engineering Dr. DY Patil Institute of Technology,Pimpri

Abstract Graphology is a method for identifying, evaluating personality traits by handwriting. Professional Handwriting analysts are called Graphologists. Handwriting is often called as Mind Writing or Brain Writing. It reflects human’s thought-process through his handwriting Accuracy of Handwriting depends upon intellectual of the Graphologists. The proposed System focuses on developing a software for predicting human behaviour. In this paper a method has been proposed from baseline, slanting of letters, looping of letters, pen pressure and height of the letters. The system uses Convolutional Neural Network (CNN) for prediction of human nature. Keywords- Behavior Prediction, Image Processing, Feature Extraction, CNN

I. INTRODUCTION Since forever, researchers, thinkers, specialists and others have been keen on the connection between the penmanship and the author. This endeavored to relate explicit penmanship components to explicit human attributes. It took some time. In 1910, Milton Newman Bunker, a shorthand educator, in Kansas, let his interest show signs of improvement of him. He needed to know why, as a handwriting understudy, he had put wide spaces between his letters and long finals on his words. He started to think about the graphology. In 1915, Bunker made his one of a kind disclosure. He perceived that every one of his understudies framed shorthand strokes in a remarkable way. He all of a sudden and plainly understood that it was not the letter which had a characteristic importance but rather the strokes – the state of the developments inside the letter. Graphology recommended that an O with an open top – that is a space opening, demonstrated an individual who might talk straightforwardly and frequently. He checked and saw this as obvious. He thought, in any case, that coherently, different letters with a similar circle development (a,g,d and q) ought to have a similar significance and in the wake of checking cautiously he found that he did subsequent to voyaging a huge number of miles, and talking a huge number of individuals and inspecting the greater part a million penmanship examples in his lifetime, the copyrighted American System of penmanship examination – Graphoanalysis was conceived. A. Objective This paper plans to anticipate human conduct through penmanship investigation. Convolutional layers apply a convolution activity to the information, passing the outcome to the following layer. The convolution copies the reaction of an individual neuron to visual boosts.

II. LITERATURE REVIEW [1]Esmeralda C. Djamal anticipated Autography development copy the composed component of every individual's periodicity and plan. By examining all basics of penmanship and translating them, utilizing regular of graphology writer could start an outline of the author's character characteristic, wistful constitution and charitable plan. In chart consistent examination, a picture is isolated into two promotion that designs properties and segment digit each character. In this examination, creator utilize graphical promotion dependent on mark and digit of character of utilization plot utilizing many-outline calculations and fake neural systems (ANN). The picture break into two space: the mark involved on nine appearance and utilization plan of letters digit space. Each space had performed preprocessing to improve the acknowledgment exactness. ANN put together classifier applies with respect to five highlights of impression which result a precision of 56-78%. While four appearance of the feeling that exposure utilizing many edge calculation result 87-100% precision.

All rights reserved by www.grdjournals.com

5


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

[2]Sandeep dhang on Handwriting Analysis of Human Behavior Based on Neural Network, Graphology or Handwriting examination is a logical technique for recognizing, assessing and comprehension of anybody character through the stroke and example uncovered by penmanship. Penmanship uncovers the genuine character including enthusiastic cost, genuineness, fears and protections and so on. Penmanship stroke mirrors the on paper draw of every individual's musicality and Style. The picture split into two territories: the mark dependent on three highlights and application type of letters digit territory. In this exploration execution assessment is finished by ascertaining mean square blunder utilizing Back Propagation Neural Network (BPNN).Human conduct is examined based on signature by utilizing neural system. [3] Javier Galbally, Julian Fierrez, Marcos Martinez-Diaz, R'ejean Plamondon E'cole Polytechnique de Montre'al center around "Quality Analysis of Dynamic Signature Based on the Sigma-Lognormal Model". In this paper creator particular that different individual morals can be accurately light up as a lot of persuasive depict sequenced together by a Markov chain. To analyze individual morals from reasonable information and to conclude individual morals over a couple of moments time, creator at that point utilize these persuasive Markov design. To guarantee the ideals of this structuring road, maker report an examination where, creator had the option to accomplish 95% exactness at anticipating vehicle driversâ€&#x; resulting activities from their beginning preliminary developments. [4] In this creator separate another conduct biometric system dependent on human PC correspondence. Creator urbanized a framework that catches the client correspondence by means of help, and uses this detectable data to confirm the distinction of a person. Utilizing systematic example credit methods, creator built up a consecutive classifier that procedures client collaboration, as announced by the client personality is viewed as genuine if a predefined precision level delivered, and the client is delegated a faker generally. Two factual models for the highlights were tried, to be specific Parsing thickness conclusion and a unimodal transfer. The framework was checked with various quantities of clients so as to evaluate the adaptability of the proposition. Exploratory outcomes show that the ordinary client correspondence with the PC by means of a pointing gadget involves social data with specific force. [5]Proposed a paper tending to issue of individual validation using signature acknowledgment is portrayed in this paper. There are two technique for confirmation: on the web and disconnected mark check. The dynamic strategies secured, depend on the examination of the shape, speed, stroke, pen weight and timing data. While the stationary strategies include general shape acknowledgment systems. The paper slanted a sharp authentic blueprint of the surviving strategies and exhibits a portion of the ongoing examination in the field. In this paper issue of selective tribute using mark observation is considered. Twain on-line and disconnected techniques have been portrayed.

III. PROPOSED METHODOLOGY Graphologist recognize human instinct with a bit of written by hand penmanship. The exactness of penmanship examination relies upon how talented the graphologist is. Despite the fact that manual penmanship has been compelling it is expensive and inclined to exhaustion, consequently the proposed procedure centers around building up a product for conduct examination which can foresee character attributes with the assistance of PC without human impedance. The for the most part utilized highlights of penmanship for expectation of character characteristics are gauge, thickness, pen pressure, stature and so on. In this paper, the standard, pen pressure, letter tallness and inclination of letters has been considered for anticipating character. Baseline: It is the line along which the writing flows.

Pen Pressure: Amount of intensity applied while writing.

Height: It is the total vertical length of letters.

Slant: It is the inclination of letters.

All rights reserved by www.grdjournals.com

6


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

Below are the implementation steps involved in Handwriting analysis.

Fig. 1: Overall Architecture

A. Image Pre-processing Picture preparing is finished with a mean to improve the picture information that smothers undesirable twists or upgrades some picture highlights significant for additional handling. The reason for picture preparing is separated into 5 gatherings. They are: 1) Perception - Observe the articles that are not obvious. 2) Picture honing and rebuilding - To make a superior picture. 3) Picture recovery - Seek for the picture of intrigue. 4) Estimation of example – Measures different articles in a picture. 5) Picture Recognition – Distinguish the items in a picture 1) Gray Scale Grayscale picture otherwise called highly contrasting picture is the one wherein every pixel of the picture conveys power data. Dim scale picture has just two hues: Black and white. The changed over grayscale picture may lose contrasts, sharpness, shadow, and structure of the shading picture. The luminance of a pixel estimation of a grayscale picture ranges from 0 to 255.

Fig. 2: Gray Scale 2) Bilateral Filter A bilateral filter is a non-linear, edge-preserving, and noise-reducing smoothing filter for images. It replaces the intensity of each pixel with a weighted average of intensity values from nearby pixels. This weight can be based on a Gaussian distribution.

All rights reserved by www.grdjournals.com

7


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

Crucially, the weights depend not only on Euclidean distance of pixels, but also on the radiometric differences (e.g., range differences, such as colour intensity, depth distance, etc.). This preserves sharp edges.

Fig. 3: Bilateral Filter

3) Canny Edges Watchful edge recognition is a used to remove valuable basic data from various articles and lessen the measure of information to be handled. The general criteria for edge recognition incorporate:1) Detection of edge with low blunder rate, which implies that the discovery ought to precisely get however many edges appeared in the picture as could reasonably be expected. 2) The edge point identified from the administrator ought to precisely confine on the focal point of the edge. 3) A given edge in the picture should just be checked once, and where conceivable, picture clamor ought not make bogus edges. To fulfill these prerequisites Canny utilized the analytics of varieties – a method which finds the capacity which upgrades a given utilitarian. The ideal capacity in Canny's indicator is portrayed by the total of four exponential terms, however it very well may be approximated by the main subordinate of a Gaussian.

Fig. 4: Canny edges 4) Contouring Molding strategy is utilized to separate data about the state of the picture. The highlights got from various attributes will be utilized in design characterization once the form of explicit example is extricated.

Fig. 5: Conturing

B. Convolutional Neural Network (CNN) Convolutional neural system (CNN, or ConvNet) is a structure profound learning and most ordinarily applied to investigating visual symbolism. CNNs utilize a variety of multilayer perceptrons intended to require negligible preprocessing. They are otherwise called move invariant or space invariant counterfeit neural systems (SIANN), in view of their common loads engineering All rights reserved by www.grdjournals.com

8


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

and interpretation invariance attributes. Convolutional systems were roused by natural procedures in that the availability design between neurons looks like the association of the creature visual cortex. Individual cortical neurons react to upgrades just in a confined district of the visual field known as the responsive field. The open fields of various neurons halfway cover with the end goal that they spread the whole visual field. CNNs utilize moderately little pre-handling contrasted with other picture characterization calculations. This implies the system learns the channels that in customary calculations were hand-designed. This freedom from earlier information and human exertion in include configuration is a significant favorable position. They have applications in picture and video acknowledgment, recommender frameworks, picture order, therapeutic picture investigation, and common language preparing. A CNN comprises of an info and a yield layer, just as various concealed layers. The concealed layers of a CNN regularly comprise of convolutional layers, pooling layers, completely associated layers and standardization layers.

Fig. 6: Simple ConvNet

The Convolutional Neural Network in Fig. is comparable in design to the first LeNet and orders an info picture into four classifications: hound, feline, pontoon or feathered creature There are four primary activities in the ConvNet appeared in fig. above: 1) Convolution 2) Non Linearity (ReLU) 3) Pooling or Sub Sampling 4) Classification (Fully Connected Layer) An Image is a grid of pixel esteems Basically, every picture can be spoken to as a grid of pixel esteems Channel is a regular term used to allude to a specific part of a picture. A picture from a standard advanced camera will have three channels – red, green and blue – you can envision those as three 2d-grids stacked over one another (one for each shading), each having pixel esteems in the range 0 to 255. 1) The Convolution Step ConvNets get their name from the "convolution" administrator. The main role of Convolution if there should be an occurrence of a ConvNet is to extricate highlights from the info picture. Convolution saves the spatial connection between pixels by learning picture highlights utilizing little squares of information. We won't delve into the scientific subtleties of Convolution here, however will attempt to see how it functions over pictures As we talked about over, each picture can be considered as a network of pixel esteems. Consider a 5 x 5 picture whose pixel esteems are just 0 and 1 (note that for a grayscale picture, pixel esteems run from 0 to 255, the green network underneath is an exceptional situation where pixel esteems are just 0 and 1):

Additionally, consider another 3 x 3 lattice as appeared. At that point, the Convolution of the 5 x 5 picture and the 3 x 3 network can be processed as appeared in the movement in Fig underneath:

Fig. 7: The Convolution operation. The output matrix is called Convolved Feature or Feature Map

All rights reserved by www.grdjournals.com

9


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

Pause for a minute to see how the calculation above is being finished. We slide the orange grid over our unique picture (green) by 1 pixel (additionally called 'walk') and for each position, we register component shrewd augmentation (between the two lattices) and add the duplication yields to get the last whole number which frames a solitary component of the yield framework (pink). Note that the 3Ă—3 network "sees" just a piece of the information picture in each walk. In CNN wording, the 3Ă—3 grid is known as a 'channel' or 'bit' or 'highlight indicator' and the lattice shaped by sliding the channel over the picture and processing the speck item is known as the 'Convolved Feature' or 'Initiation Map' or the 'Element Map'. Note that channels goes about as highlight finders from the first info picture. It is apparent from the liveliness over that various estimations of the channel framework will create diverse Feature Maps for a similar info picture. For instance, consider the accompanying info picture: In the table underneath, we can see the impacts of convolution of the above picture with various channels. As appeared, we can perform tasks, for example, Edge Detection, Sharpen and Blur just by changing the numeric estimations of our channel network before the convolution activity this implies various channels can recognize various highlights from a picture, for instance edges, bends and so forth.

2) Presenting Non Linearity (ReLU) An extra activity called ReLU has been utilized after each Convolution activity in Figure above. ReLU represents Rectified Linear Unit and is a non-direct activity. Its yield is given by:

Fig. 8: The ReLU operation

All rights reserved by www.grdjournals.com

10


Behaviour Analysis using Handwritten Data (GRDJE/ Volume 5 / Issue 3 / 002)

ReLU is a component astute activity (applied per pixel) and replaces all negative pixel esteems in the element map by zero. The motivation behind ReLU is to present non-linearity in our ConvNet, since a large portion of this present reality information we would need our ConvNet to learn would be non-direct (Convolution is a straight activity – component shrewd grid duplication and expansion, so we represent non-linearity by presenting a non-direct capacity like ReLU). 3) The Pooling Step Spatial Pooling (likewise called subsampling or downsampling) diminishes the dimensionality of each component map however holds the most significant data. Spatial Pooling can be of various kinds: Max, Average, Sum and so forth. If there should be an occurrence of Max Pooling, we characterize a spatial neighborhood (for instance, a 2×2 window) and take the biggest component from the corrected element map inside that window. Rather than taking the biggest component we could likewise take the normal (Average Pooling) or total of all components in that window. By and by, Max Pooling has been appeared to work better. Shows a case of Max Pooling activity on a Rectified Feature map (got after convolution + ReLU activity) by utilizing a 2×2 window.

Fig. 9: Max Pooling

We slide our 2 x 2 window by 2 cells (additionally called 'walk') and take the most extreme incentive in every locale. As appeared in Figure, this diminishes the dimensionality of our component map. In the system appeared in fig.

IV. CONCLUSIONS A less difficult strategy has been proposed to foresee the character of an individual by investigating his penmanship. The framework extricates highlights from breaks, size, space between words, gauge, circle of 'e' and scarcely any different highlights like weight, edge, inclination and spot separation in 'I'. The proposed framework can be utilized as a twin device by graphologist to improve the precision and envision the conduct s of an individual quicker. The assessed weighted exactness of 93.77 % is accomplished.

REFERENCES Champa H N, Dr. K R Ananda Kumar, “Artificial neural network for human behaviour prediction through handwriting analysis”, International Journal of Computer Applications (0975 – 8887) Volume 2-No.2, May 2010. [2] “Automated Human behavior prediction through handwriting analysis” By Champa H N and K R Anandakumar [3] S. M. E. Hossain and G. Chetty, ”Human Identity Verification by Using Physiological and Behavioral Biometric Traits”, International Journal Bioscience, Biochemistry and Bioinformatics, Vol. 1, No.3September 2011 [4] Esmeralda C Djamal, Sheldy Nur Ramdlan, Jeri Saputra, “Recognition of Handwriting Based on Signature and Digit of Character Using Multiple of Artificial Neural Networks in Personality Identification”, Information Systems International Conference (ISICO), 2 – 4 December 2013 [5] Sandeep Dang, Prof. Mahesh Kumar, Mahesh, “Handwriting Analysis of Human Behaviour Based on Neural Network”, ISSN: 2277 128X, Volume 4, Issue 9, September 2014 [6] Javier Galbally, Julian Fierrez, Marcos Martinez-Diaz, R´ejean Plamondon E´cole Polytechnique de Montre´al,”Quality Analysis of Dynamic Signature Based on the Sigma- Lognormal Model”, 2011 International Conference on Document Analysis and Recognition [7] Albert Ali Salah1, Theo Gevers1, Nicu Sebe2, and Alessandro Vinciarelli3,” Challenges of Human Behavior Understanding” [8] Handwriting Personality Profile. http://handwritingpro.com (1998) [9] S. M. E. Hossain and G. Chetty, ”Human Identity Verification by Using Physiological and Behavioral Biometric Traits”, International Journal of Bioscience, Biochemistry and Bioinformatics, Vol. 1, No. 3, September 2011 [10] Hugo Gamboaa and Ana Fredb,” A Behavioral Biometric System Based on Human Computer Interaction” [11] Ajzen I, Fishbein M. (1999) Theory of reasoned action/Theory of planned behavior. University of South Florida. [1]

All rights reserved by www.grdjournals.com

11


Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.