You are on page 1of 14

Pattern Recognition

Contents

1.Introduction
2.Object Recognition
3.Approaches for Object Recognition
- Appearance Based Methods
- Feature Based Methods
4.

Introduction
What is Digital
Image?

A digital image is a representation of a 2dimensional as a finite set of digital values, for


picture elements or pixels.
Pixel values typically represents grey levels, colors,
heights, opacities etc.

What is Digital Image


Processing?

Digital image processing focusses on two major tasks :


1. Improvement of pictorial representation for human
interpretation.
2. Processing of image data for storage, transmission
and representation or machine perception.

Object Recognition
Object recognition is a process for identifying a
specific object in a digital image or video. Object
recognition algorithms rely on matching, learning, or
pattern recognition algorithms using appearancebased or feature-based techniques.

Approaches for Object Recognition


Appearance
Based
Methods:
This
technique
incorporates templates of objects to perform recognition.
It further involves further methods for the recognition as
the objects look different under various conditions such as
changes in lighting or color, viewing direction, size-shape
etc. The methods involved in recognition are:
1.
2.
3.
4.

Edge Matching
Divide-and-Conquer Search
Greyscale Matching
Histogram of receptive field responses

1. Edge Matching: It uses certain image detection


techniques to find the edges as variation in lighting
and color do not have much effect on the image edges.
2. Divide-and-Conquer Search: This approach relies
on finding a lower bound on the best score, for an
image, if bound is found to be too large , get rid of it. If
bound is within acceptable range then this cell is
further divided into cells, thus repeating this process,
until cell is small enough.
3. Greyscale Matching: Edges are robust to
illumination changes. For matching images pixel
distance is computed as a function of pixel position
and pixel intensity.
4. Histogram of Receptive Field Responses:
Relations between different image points implicitly
coded in the receptive field responses

Feature Based Methods : This technique


involves features which are used to search for
feasible matches between object features and
image features. Various features of an image like
surface patches, corners, linear edges etc. are
accounted for the identification of the object.
Some of the methods involved in feature based
methods are :
1. Hypothesize and test
2. Scale-Invariant feature transform

Scale-Invariant Feature Transform:


-Can identify objects even among clutter and under
partial occlusion.
-SIFTs feature descriptor is invariant to uniform
scaling, orientation and partially invariant to
distortion and illumination changes.
-Key points of objects are first extracted from a set
of reference images and stored in database.
-An object is recognized in a new image by
individually comparing features from the new image
to the database.

Flowchart for Object


Recognition
Detect
Initialize the
Take 1st picture

camera

predefined
object

Object
Recognized

Obtain object
properties

Yes
Do the
properti
es
match?
Compare
properties of
1st and 2nd
image

No

Take
another
picture?

Obtain object
properties

Yes
Take 2nd
picture

Detect
predefined
object

Load Image

No. of circles in the


given image are to be
found out using MATLAB
functions.

Initial attempt to find the circle

The initially uploaded


image is converted to
Greyscale using MATLAB
function:

After the conversion the


centre coordinates and
the radii of the circles
are found out.

Red circles drawn on the image over the


identified shapes

Total no. of circles found out: 16

Pattern Recognition
Pattern recognition is the process of classifying input
data into objects or classes based on key features.
There are two classification methods in pattern
recognition:
supervised
and
unsupervised
classification.
Pattern recognition algorithms generally aim to
provide a reasonable answer for all possible inputs
and to perform "most likely" matching of the inputs,
taking into account their statistical variation.
Pattern recognition has applications in computer
vision, radar processing, speech recognition and text
classification.

Future Aspects
Extensive research on various methods and
algorithms regarding object identification is being
carried out to increase the flexibility and
efficiency of the identification of the objects.
Trying out various algorithms and commenting on
their efficiency with respect to time and accuracy.
Identification of various geometric shapes
invariant of their size and orientation.
Identification of various components in a given
circuit board/

You might also like