Vous êtes sur la page 1sur 28

C o n te n t b a se d I a g e m R e tri va l e

Project By:Vivian Emmanuel (07bec020) Krishna Mewawala (07bec047) Roopam Agrawal (07bec132)

Guided By:Prof. Tanish Zaveri

Introduction

Aim

- To analyze the microscopic images and identify the plant based on its features using Content Based Image Retrieval technique This project is in collaboration with Institute of Pharmacy Presently, the job of identification of the plant is being done manually Useful in pharmaceutical companies

Project Overview
Provided

with the database of microscopic images of Vasaka, Kurchi, Rhubarb, Licorice, Datura Designed an algorithm for detection of plants Prepared a GUI for the whole process which will help in abstraction and encapsulation of the key processes involved in analysis Worked towards the optimization of the solution and increasing efficiency.

Introduction to Plants

Dhatura

Vasaka

Licorice

Rhubarb

Kurchi

Algorithm for detection

GUI

Result Statistics
Vasaka Dhatura Rubarb Licorice Kurchi

64.75% 65%

75% 77.77%

70%

Improved Algorithm
Fast

shape detection methods and improvement in shape detection algorithm Different options in GUI itself for fast detection of plant Minimum wrong detection

Fast detection options


Detection

by part- leaf, root etc. Detection by nearest guess. Detection by zooming type(10X, 45X etc.)

Circular shape detection


Hough Transform

For Circle Detection with unknown radius Circle Hough Transform(FCHT)

Fast

Hough Transform For Circle Detection - with unknown radius


For

circle equation is r2 = (x-a)2+(y-b)2

Converting

it into polar coordinates x = a+rcos() y = b+rsin()

Hough Transform For Circle Detection - with unknown radius(cont.)

Fast Circle Hough Transform (FCHT)


strategy Manages to reduce the execution times Important for.. 1) Multiple figures appear in the image 2) Figures of very different sizes
New

Fast Circle Hough Transform (FCHT) (cont.)


Focusing Applied

Algorithm

in 2-D

square

parameter space is divided into four quadrants. A given quadrant is divided again..

Fast Circle Hough Transform (FCHT) (cont.)


Detection of Centre(using gradient)

Detection of Radius

Example

Active Nodes-8

Statistics
Active Nodes 4 8 12 16 Time(in sec) 2.3 2.0 1.8 1.5 Speed up(in %) 0 13 21 34

Genetic Algorithm
Step

1: Detecting regular shapes (lines, circles, parabola, ellipses) Step II: Elimination of detected shapes from image Step III: Initialization of chromosomes which felicitates fast detection

Pattern Detection
Parameters

used to define basic shapes:

rx, ry, cx, cy,


The

matching ratio is defined as:

Fuzzy C Quadric Shells algorithm


Conventional

method used Division of the image into clusters Individual shape detection applied at each cluster Calculate the center and shape of each cluster w.r.t. some reference Iteration of the above process for a fixed number of times

Flow chart depicting process

Nagao et al. (Partial Shape Vectors) Can detect partially

Nagao et al. (Genetic algorithm)

occluded shapes accurately Robust against noise However limited shapes can be detected

Can

detect a shape even if it is rotated, shifted, reduced or enlarged Only one shape per process Range of shapes limited

Circular Hough Transform in three dimensions (x,y,z)

Fast Circle Hough Transform in only two dimensions

Works

Works

Slower

Faster

Easy

Algorithm

Complicated

COMPARISON

Future work
Apply

above algorithm(for circle detection) using matlab Find out other fast detection methods for detection of ellipses, lines and irregular shapes Create database for more plants Optimise the solution for maximum efficiency.

ACKNOWLEDGEMENT
It

gives us a great pleasure in expressing thanks and profound gratitude to Prof. Tanish Zaveri, Department of Electronics and communication Engineering, Institute of Technology, Nirma University of Science and Technology, Ahmedabad for his great valuable guidance and continual encouragement throughout the seminar.

Special Thanks

Dr. Sanjay Acharya and Nirma Institute of Pharmacy for providing us with data-base and guidance regarding it.

Thank You

Vous aimerez peut-être aussi