International Journal of Engineering Trends and Technology- Volume4Issue3- 2013

IMAGE STITCHING USING MATLAB

Tejasha Patil, Shweta Mishra ,Poorva Chaudhari , Shalaka Khandale
Information Tech. Department
P.V.P.P.C.O.E.
Mumbai, India

Abstract
Images are an integral part of our daily lives. Image stitching
is the process performed to generate one panoramic image
from a series of smaller, overlapping images. Stitched images
are used in applications such as interactive panoramic viewing
of images, architectural walk-through, multi-node movies and
other applications associated with modeling the 3D
environment using images acquired from the real world.
Image processing is any form of signal
processing for which the input is an image, such as a
photograph or video frame; the output of image processing
may be either an image or, a set of characteristics or
parameters related to the image. Most image processing
techniques involve treating the image as a two-dimensional
signal and applying standard signal processing techniques to
it. Specifically, image stitching presents different stages to
render two or more overlapping images into a seamless
stitched image, from the detection of features to blending in a
final image.
In this process, Scale Invariant Feature
Transform (SIFT) algorithm[1] can be applied to perform
the detection and matching control points step, due to its good
properties. The process of create an automatic and effective
whole stitching process leads to analyze different methods of
the stitching stages. Several commercial and online software
tools are available to perform the stitching process, offering
diverse options in different situations.
Key words : seamless stitched image, Panoramic image.
1. Introduction
This projects goal is to create a Matlab script that will stitch
two images together to create one larger image. Given a
sequence of images taken from a single point in space, but
with varying orientations, it is possible to map the images into
a common reference frame and create a perfectly aligned
larger photograph with a wider eld of view. This is normally
referred to as panoramic
image stitching.
In our experiments, we observe that a successful image
stitching algorithm should not only create a smooth transition
within the overlapped region but also preserve the following

ISSN: 2231-5381

properties, which are in general agreement with our visual

perception: Structure preservation. The stitched image should
not break existing or create new salient structures. where the
edge of the tower is broken in the overlapped region due to
structure misalignment, causing obvious ghosting artifact.
Intensity alignment. Human eyes are sensitive to large
intensity change. Unbalanced contrast beyond the overlapped
area of a stitched image can be perceptually magnified.
Although the structure is well aligned and color transition is
smooth within the overlapped area, the unnatural color
transition from left to right reveals the unmatched intensities
inherent in the input images. The context information of
objects in the input images should be taken into account
during the stitching process.
2. CONTEXTUALIZATION
2.1 Existing sysem
Due to the limited Field-Of-View (FOV) of a single camera, it
is sometimes desired to extend the FOV using multiple
cameras. Image stitching is one of the methods that can be
used to exploit and remove the redundancy created by the
overlapping FOV. However, the memory requirement and the
amount of computation for conventional implementation of
image stitching are very high. In this project, this problem is
resolved by performing the image stitching and compression
in a strip-by-strip manner. First, the stitching parameters are
determined by transmitting two reference images to an
intermediate node to perform the processing. Then, these
parameters are transmitted back to the visual node and stored
in there. These parameters will be used to determine the way
of stitching the incoming images in a strip-by-strip manner.
After the stitching of a strip is done, it can be further
compressed using a strip-based compression technique.
Most of the existing methods of image stitching either
produce a rough stitch that cannot deal with common
features such as blood vessels, comet cells and histology, or
they require some user input. Approaches for image stitching
that optimize the search for the best correlation point by using
Levenberg-Marquardt method .Levenberg-Marquardt method
gives good results, but it is computationally expensive and can
get stuck at local minima. The approach offered in this project
makes the selection of the best correlation point in the
following way. Based on knowledge about the expected
overlap when using the motorized stage, it would be

http://www.internationaljournalssrg.org

Page 302

International Journal of Engineering Trends and Technology- Volume4Issue3- 2013

straightforward to find the best correlation point in the ideal
case. However, the overlap area is not perfect, and certainly
not to an accuracy of one pixel, due to deviations in stage
position from the ideal and due to stage/camera misalignment.
Our algorithm offers a way to overcome this problem by
searching the small area around the expected central overlap
pixel in order to find the best correlation point. Positioning of
acquired images with a manual stage is much less accurate, so
there is a need to search a wider area in order to find the best
cross-correlation point.
This projects goal is to create a Matlab script that will stitch
two images together to create one larger image. Image
stitching has wide uses in photo applications and has become a
required toolset for many photographers. These stitched
images, become panoramic views which increase the visual
aesthetics of a scene, and are widely sought out for posters,
postcards, and other printed materials. This project will be
performed using point correspondences between the two
images and utilizing Matlabs Image Processing Toolbox.
2.2 Proposed System
Image processing can be defined as analysis of picture using
techniques that can basically identify shades and colors. It
deals with images in bitmapped graphic format that have been
scanned or captured with digital camera. It also means image
improvement, such as refining a picture in a program or
software that has been scanned or entered from a video source
or in short, image processing is any form of information
processing when both the input and output is images. Image
processing is divided into two major branches; image
enhancement and image restoration. Image enhancement is to
improve the quality of image or emphasize particular aspects
within image and to produce image that is different from the
original. Whereas image restoration, is to recover the original
image after degraded by known effects such as geometric
distortion within camera system. Image processing does not
reduce the amount of data present but rearranges it.
Spatial filtering :
An image can be filtered to remove a band of spatial
frequencies, such as high frequencies and low frequencies.
Where rapid brightness transitions are established, high
frequencies will be there. In the other hand, slowly changing
brightness transitions represent low frequencies. The highest
frequencies normally will be found at the sharp edges or
points.
Spatial filtering operations include high pass, low pass and
edge detection filters. High pass filters accentuate the high
frequency details of image and attenuate the low frequency,
creating a sharpening effect.
Sharpening :
The main aim in image sharpening is to highlight fine detail in
the image, or to enhance detail that has been blurred due to
noise or other effects. Sharpening emphasizes edges in the
image and make them easier to see and recognize. In creating
a sharpened image, no new details are actually created. The
nature of sharpening is is influenced by the blurring radius
used. In addition to that, differences between each pixel and
its neigbour too, can influence sharpening effect.
Blurring :

ISSN: 2231-5381

The visual effect of a low pass filter is image blurring. This is

because the sharp brightness transitions had been attenuated to
small brightness transitions. It have less detail and blurry.
Blurring can be done in spatial domain by pixel averaging in a
neighbor. Blurring is aimed to diminish the effects of camera
noise, unauthentic pixel values or missing pixel values. For
blurring effect, there are two mostly used techniques which
are neighbourhood averaging (Gaussian filters) and edge
preserving (median filters). The blurring effect can improve
an images low frequency details by removing visually
disruptive high frequency patterns. By subtracting a low pass
filtered image from the original image, a sharpened image can
be created. The latter operation is known as unsharp masking
enhancement.
Edge detection :
Edges are often used in image analysis for finding region
boundaries. They are pixels where brightness changes
abruptly. An edge essentially distinguishes between two
distinctly different regions or in short, an edge is the border
between two different regions. Robert operator, Sobel
operator, Prewitt operator, Canny operator and Krish operator
are among edge detectors that are often used. Edge detection
of an image reduces significantly the amount of data and
filters out information that may be regarded as less relevant,
preserving the important structural properties of an image.
There are many methods for edge detection, but most of them
can be grouped into two categories, search-based and zerocrossing based. The 3 important stages in the process of
stitching images are:
1.Image Acquisition:
The images that need to be stitched are acquired using a
camera mounted on a tripod stand. The camera angle is varied
to take different overlapping sample images.
2.Image Registration:
The process of image registration aims to find the translations
to align two or more overlapping images such that the
projection from the view point through any position in the
aligned images into the 3D world is unique.
Image Registration consists of four main components:
Feature set- The set of features includes the intensity values,
contours, textures and so on. A feature set must be selected for
each image registration method.
Similarity measure- It is a function which returns a scalar
value that provides an indication of the similarities between
two features.
Search set- It is a set of possible transformations for aligning
the images.
Search strategy- It is the algorithm that decides how to select
the next transformations from the search set.
Techniques used:
1) Registration using a different feature set.
2) Registration using different similarity measures.
3) Registration with step search strategy.
3) Image Merging:
Image merging is the process of adjusting the values of pixels
in two registered images, such that when the images are
joined, the transition from one image to the next is invisible. It
should also ensure that the new image has a quality

http://www.internationaljournalssrg.org

Page 303

International Journal of Engineering Trends and Technology- Volume4Issue3- 2013

comparable to that of the original images used. Image merging
can be carried out by making the seam invisible in the output
image. The seam is the line that is visible at the point where
the two images overlap.
Techniques used:
1) Linear distribution of intensity differences
2) Linear distribution of median intensity differences
3) Intensity adjustment with respect to corresponding
pixels in overlapping region.
Steps for Image Sitching :

the purpose of object recognition is proposed. The main idea is

to divide the features extracted from both the test and the
model object image into several sub-collections before they
are matched. The features are divided into several subcollections considering the features arising from different
octaves, that is from different frequency domains. [1]
Following are the major stages of computation used to
generate the set of image features:
1. Scale-space extrema detection: The rst stage of
computation searches over all scales and image locations. It is
implemented efciently by using a difference-of-Gaussian
function to identify potential interest points that are invariant
to scale and orientation.
2. Keypoint localization: At each candidate location, a detailed
model is t to determine location and scale. Keypoints are
selected based on measures of their stability.
3. Orientation assignment: One or more orientations are
assigned to each keypoint location based on local image
gradient directions. All future operations are performed on
image data that has been transformed relative to the assigned
orientation, scale, and location for each feature, thereby
providing invariance to these transformations.
4. Keypoint descriptor: The local image gradients are
measured at the selected scale in the region around each
keypoint. These are transformed into a representation that
allows for signicant levels of local shape distortion and
change in illumination.
This approach has been named the Scale Invariant Feature
Transform (SIFT), as it transforms image data into scaleinvariant coordinates relative to local features. [4]
RANdom SAmple Consensus algorithm
The Random sample Consensus algorithm (RANSAC)
proposed by Fischler and Bolles is a general parameter
estimation approach designed to cope with a large proportion
of outliers in the input data. Unlike many of the common
robust estimation techniques such as M-estimators and leastmedian squares that have been adopted by the computer vision
community from the statistics literature, RANSAC was
developed from within the computer vision community.
RANSAC is a resampling technique that generates candidate
solutions by using the minimum number observations (data
points) required to estimate the underlying model parameters.
As pointed out by Fischler and Bolles , unlike conventional
sampling techniques that use as much of the data as possible to
obtain an initial solution and then proceed to prune outliers,
RANSAC uses the smallest set possible and proceeds to
enlarge this set with consistent data points.
The basic algorithm is summarized as follows:

3. Methodology
Scale Invarient Feature Algorithm
The SIFT algorithm (Scale Invariant Feature Transform)
proposed by Lowe is an approach for extracting distinctive
invariant features from images. It has been successfully
applied to a variety of computer vision problems based on
feature matching including object recognition, pose
estimation, image retrieval and many others. However, in realworld applications there is still a need for improvement of the
algorithms robustness with respect to the correct matching of
SIFT features. In this paper, an improvement of the original
SIFT algorithm providing more reliable feature matching for

ISSN: 2231-5381

1) Select randomly the minimum number of points required to

determine the model parameters.
2) Solve for the parameters of the model.
3) Determine how many points from the set of all points t
with a predened tolerance  .
4) If the fraction of the number of inliers over the total number
points in the set exceeds a predened threshold , re-estimate
the model parameters using all the identied inliers and
terminate.
5) Otherwise, repeat steps 1 through 4 (maximum of N times).

http://www.internationaljournalssrg.org

Page 304

International Journal of Engineering Trends and Technology- Volume4Issue3- 2013

An advantage of RANSAC is its ability to do robust
estimation of the model parameters, i.e., it can estimate the
parameters with a high degree of accuracy even when a
significant number of outliers are present in the data set. A
disadvantage of RANSAC is that there is no upper bound on
the time it takes to compute these parameters. When the
number of iterations computed is limited the solution obtained
may not be optimal, and it may not even be one that fits the
data in a good way. In this way RANSAC offers a trade-off;
by computing a greater number of iterations the probability of
a reasonable model being produced is increased. Another
disadvantage of RANSAC is that it requires the setting of
problem-specific thresholds.
RANSAC can only estimate one model for a particular data
set. As for any one-model approach when two (or more)
model instances exist, RANSAC may fail to find either one.
The Hough transform is an alternative robust estimation
technique that may be useful when more than one model
instance is present. [3]. The figure below depicts keypoint
detection.
4. CONCLUSION
In this paper, we have proposed a novel image-matching
algorithm. Our algorithm can significantly increase the
number of matching pairs, and matching accuracy. Extensive
experimental results show that our method improves the
traditional detectors, even in large variations, and the new
indicators are distinctive.
The image stitcher provides a cost effective and
flexible alternative to acquire panoramic images using a
panoramic camera. The panoramic images stitched by a
stitcher can also be used in applications where the camera is
unable to obtain a full view of the object of interest. The full
view of the object can be constructed using the image stitcher
using overlapping regional images acquired for the object.
5. ACKNOWLEDGEMENT
This paper describes research done at PVPPCOE in the
Department of Information Technology. We are thankful
to Mrs. Sonali Pakhmode for guiding us.

3. PRELIMINARY RESULTS
In this paper, we have proposed a novel image stitching
approach by image deformation, where the overlapped regions
may contain significant intensity inconsistency and
geometrical misalignment. The below example depicts two
geometrically misaligned images stitched in a panorama using
MATLAB.

6. References
[1] Y. Yu, K. Huang, and T. Tan, A Harris-like scale
invariant feature detector, in Proc. Asian Conf. Comput. Vis.,
2009, pp. 586595.
[2] J. M. Morel and G. Yu, Asift: A new framework for fully
affine invariant image comparison, SIAM J. Imag. Sci., vol.
2, no. 2, pp.
438469, Apr. 2009.
[3] J. Rabin, J. Delon, Y. Gousseau, and L. Moisan,
RANSAC: A robust algorithm for the recognition of multiple
objects, in Proc. 3DPVT, 2010.
[4] M. Krulakova, Matrix technique of image processing in
Matlab, ICCC'2010: proceedings of 11th International
Carpathian Control Conference, 26-28 May, 2010, Eger,
Hungary, Rekatel 2010, pp. 203-206, ISBN 978-963-06-92892.

ISSN: 2231-5381

http://www.internationaljournalssrg.org

Page 305

International Journal of Engineering Trends and Technology- Volume4Issue3- 2013

[5] Wei Xu and Jane Mulligan. Performance evaluation of
color correction approaches for automatic multi-view image
stitching. In 2010 IEEE Conference on Computer Vision and
Pattern Recognition (CVPR 2010), pages 263 - 270, San
Francisco, CA, USA, June 2010.
[6] Oliver Whyte, Josef Sivic1, Andrew Zisserman, and Jean
Ponce. Nonuniform deblurring for shaken images. In 2010
IEEE Conference on Computer Vision and Pattern
Recognition (CVPR 2010), pages 491 - 498, San Francisco,
CA, USA, June 2010.
[7] Xianyong Fang, Bin Luo, Haifeng Zhao, Jin Tang, Biao
He, and Hao Wu. A new multi-resolution image stitching with
local and global alignment. IET Computer Vision, 2010.
[8] MathWorks, MATLAB Builder JA 2 user's guide. [online]
August 18, 2010 [cited 12.01.2011] available from
http://www.mathworks.com/help/pdf-doc/
javabuilder/javabuilder.pdf.
[9] MathWorks, Bringing java classes and methods into
MATLAB workspace. [online] [cited 12.01.2011] available
from http://www.mathworks.com/help/ techdoc/matlabexternal/f4863.html
[10] Chen Hui, Long AiQun, Peng YuHua. Building
Panoramas from Photographs Taken with An Uncalibrated
Hand-Held Camera. Chinese Journal of Computers,
2009,(2):328-335.
[11] Hsieh, J.-W. Fast stitching algorithm for moving object
detection
and mosaic construction. in IEEE International Conference on
Multimedia & Expo. 2003. Baltimore, Maryland, USA.

ISSN: 2231-5381

http://www.internationaljournalssrg.org

Page 306