Presentation is loading. Please wait.

Presentation is loading. Please wait.

CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching Link: singhashwini.mesinghashwini.me.

Similar presentations


Presentation on theme: "CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching Link: singhashwini.mesinghashwini.me."— Presentation transcript:

1 CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Singhars1546@rit.edu Imran Bohariiib8085@rit.edu Image Stitching Link: singhashwini.mesinghashwini.me

2  Scope  Introduction  Image Stitching Approaches  Image Stitching Model  SIFT  RANSAC  Challenges  Future Scope Overview

3 To develop an image stitching algorithm which will create a panoramic image from a set of images by using feature extraction method Scope

4  Image stitching is also known as mosaicking  It is a process of combining multiple pictures of the same scene into a single high resolution image  There are two techniques of achieving image stitching:  Direct Technique  Feature Based Technique Introduction

5  It compares all pixels in the images with each other  Advantages:  They make optimal use of information available in image alignment  They measure contribution of every pixel in the images  Disadvantages:  Computationally complex  Time consuming  Examples:  Fourier Analysis Technique  A Unifying Framework Direct Technique

6  It aims to determine similarity between images by comparing features  They establish correspondence between lines, edges, corners, points and other geometric entities  Steps in these techniques:  Feature Extraction  Registration  Blending  Advantages: Robust against scene movement, faster, ability to detect panoramas etc  Examples: SIFT, SURF, ORB etc Feature Based Technique

7 Image Stitching Model Image Acquisition Feature Detection and Matching Image Matching Global Alignment Blending

8  Image Acquisition : it involves retrieving an image from the source  Feature Detection & Matching : it compares features between images in order to determine translation  Image Matching : it finds which image is neighbor of other image  Global Alignment : the aim is to find a set of parameters which minimize mis-registration between images  Blending : it selects final surface for the image and blends the images to form a single smooth image Image Stitching Model

9 The SIFT technique is one of the most robust and widely used image matching algorithm based on local features. SIFT is a feature detection and description technique. SIFT produces key point descriptors which describes the image features.

10 Construct Scale space DoG Locate DoG Extrema Potential feature points Filter edge Assign key points orientations Build key point descriptors Store the vector descriptor Do the matching

11 Gaussian Kernel is used to create Scale space. Down sample every time by a factor of 2. Take the difference of last Gaussian and store it.

12 Scan all DoG images obtained from previous process. Take top 2 images from the stack. Look at all neighboring points and decide the maxima and minima. For each pixel, marked as X there will be 26 points to compare.

13

14 Find the orientation of each feasible potential points. Calculate gradient dx and dy. Calculate direction theta. For each 16x16 block, we register direction in 8 different views.

15 Random sample consensus ( RANSAC ) is an iterative method to estimate parameters of a mathematical model from a set of observed data which contains outliers. RANSAC Algorithm Repeat N times 1.Randomly select a sample Select just enough points to recover the parameters 2.Fit the model with random sample 3.See how many other points agree Best estimate is one with most agreement

16 http://6.869.csail.mit.edu/fa12/lectures/lecture13ransac/lecture13ransac.pdf

17

18

19

20

21

22 Initial Matched Points https://courses.engr.illinois.edu/cs498dwh/fa2010/lectures/Lecture%2017%20-%20Photo%20Stitching.pdf

23 Final Matched Points https://courses.engr.illinois.edu/cs498dwh/fa2010/lectures/Lecture%2017%20-%20Photo%20Stitching.pdf

24

25

26

27  An image is often corrupted with noise during its acquisition which in turn hampers feature extraction  Image stitching on large set of data requires more time  Small scene motion such as waving tree branch and large scene motion like people moving in and out of the pictures  Images must be taken from the same spot i.e. without moving the camera Challenges

28  Photosynth: stitching many images into one large composite image  Video Stitching Future Scope

29  Ebtsam Adel Mohammad Elmogy Hazem Elbakry. “Image Stitching Based on Feature Extraction Techniques: A survey”. International Journal of Computer Applications, volume 99, August 2014.  Shikha Arya. “A Review on Image Stitching and its Different Methods”. International Journal of Advanced Research in Computer Science and Software Engineering, Volume-5, Issue-5, May 2015.  Image Stitching Wikipedia: https://en.wikipedia.org/wiki/Image_stitching References

30 Please feel free to ask any questions or concerns THANK YOU Questions & Discussion


Download ppt "CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching Link: singhashwini.mesinghashwini.me."

Similar presentations


Ads by Google