You are on page 1of 3

Signature Recognition

Signature Recognition is type of image recognition is done in


two steps those are Feature detection and Feature recognition.
These steps can be achieved by extracting features from the
training data source this can be accomplish by Feature
extraction.Feature Extraction is a dimensionality
reduction process, where an initial set of raw variables is reduced to
more manageable groups (features) for processing, while still
accurately and completely describing the original data.
Feature recognition is the process to classify the extracted
features and recognise with the saved features. The process of the
signature verification we will use the Oriented FAST and Rotated
BRIEF algorithms to extract the features of the signatures and
FLANN is a library for quickly estimating nearest neighbors. This
technique consists of various optimization algorithms for searching
fast neighbors and high-dimensional features and also we will try to
implement the other feature algorithms like SIFT, SURF etc. At last
we might be using a mean measure across these algorithms for
better accuracy.
The detailed description of the algorithms that we will be using
for the signature recognition is given below
The ORB algorithm does the FAST key point detection and
BRIEF( Binary Robust Independent Elementary Features) feature
extraction. It uses oFAST algorithm to detect the position of key
points. The oFAST, is the FAST(Features from Accelerated
Segment Test) orientation, increasing the calculation of key points
in the direction of the corner. Since multiple images need to be
processed the original image is pyramided then it uses the FAST
algorithm to find the location of the key points. Harris corner
detection selects the top N best points. These are processed for
each image. For each corner point, it calculates the direction of the
corner points according to the Intensity Centroid algorithm. rBRIEF,
Rotation-Sensitive-BRIEF, The brief algorithm is undirected. So the
direction of the corner calculated using Intensity Centroid algorithm
as the direction of the brief and rotates to get the directed brief, that
is, the steered brief. The steered brief results in a decrease in
performance. The value of each bit of the brief feature is either 0 or
1. After the steered brief feature is added to the direction of the
corner point, the distribution of the feature mean is even. If all the
features are similar and closed to mean it is not easy to distinguish.
The ORB feature is a combination of oFAST and rBRIEF
implementing the ORB algorithm, relevant algorithms, FAST, Harris
corners, Intensity Centroid, BRIEF, etc., are involved, these
featuers are stored in the database. Here will use OpenCV
implementation of orb for the detection purpose across the input
and stored db.
We use Flann, Fast Library for Approximate Nearest
Neighbors to classify and features of the signature. Dictionaries are
used to store the calculated descriptors. If the matched value is less
than the threshold after comparing with the original signature(s) in
the database then it is said to be a forgery,. Similarly Based on the
maximum number of Descriptors matches we can determine the
original signature.
SIFT Scale-invariant feature transform descriptor for the feature
extraction and key points are selected using k-nearest neighbor
rule.
Below one is the basic architecture for Signature Recognization

You might also like