# Abstract Video summarization is a process of creating

Abstract

A face Recognition method is

implemented to find a particular person. Here we are detected specific player

from whole cricket inning video and summarize it, there are many techniques are

available for face detection, in this paper we are discussed some of them in

brief.

Keywords

Object

Detect Using SURF, Face Recognition Using PCA, LDA

Introduction

Video summarization is a process of creating &

presenting a meaningful abstract view of entire video within a short period of time.

Main objective of this paper is to summarize whole video only for Specific

Player. Abstract video for specific shorts like four, six, out etc. and high

light the specific event from video.

Literature Review

In 2015, Jageshvar K. KecheVikas K.

YeotikarManish T. WanjariDr. Mahendra P. Dhore, presented a document on the

“Recognition of human faces based on the PCA method using MATLAB”.

The system receives input from the ENT database and is recognized by the training

set. Recognition is done by finding the Euclidean distance between the entrance

face and our training set. The results were simulated using MATLAB. This

approach is certainly simple, easy and fast to implement identification,

verification and authentication 3.

In 2016, A. Al-Asadi Tawfik and Ahmed

Obaid J., the document “Detection and object using an improved and

accelerated improvement of the function” present technique based on the

detection of objects. to detect and recognize objects in the scene is based on

the SURF algorithm, improving the detection performance of object descriptor

selecting stronger features, our proposed method correctly detects one or more

objects in the image data set and calculates the corresponding scores to the

object in the scene by applying three types of threshold and precision

measurements are the recognition of objects in different conditions of

rotation, partial occlusion, changes in lighting improved illumination image

input and orientation. Many real-time applications using the SURF algorithm can

detect how objects are displayed, our model calculates a lot of information

that is used throughout the detection of phase objects, so our proposed model

is easy to use, in which it is possible to select and change many parameters

such as the selected threshold and the octaves used for the detection and

recognition process 2.

In 2016, Yukti Bakhshi1, Sukhvir Kaur2 and

Prince Verma3, his work “An Effective Approach for Facial Recognition

Faces Using SIFT, SURF and PCA” Present three methods are SIFT, SURF and

PCA for face recognition to solve the coincidence problem of images in the case

of invariable faces. This method is quick and offers better recognition speed.

This is a method of cash facial recognition using the SIFT and SURF features to

extract the characteristics of facial images and, finally, the PCA technique is

applied to the image to achieve better results in case of variations in

expression and contrast, as well as the rotation. The local PCA descriptors

SIFT and SURF are more robust than the original local SIFT descriptors 4.

In 2009, Geng Du *, Fei Do, Cai Anni,

presented a report on “Face Recognition using Surf features”. They

present the features of SURF in facial recognition and offer detailed

comparisons with the characteristics of SIFT. The features of SURF have

slightly better performance than SIFT, but there is a clear improvement in the

speed of coincidence. Therefore, the characteristics of SURF proved to be

adequate for facial recognition.

In 2012, Muhammad Ajmal, Muhammad Ashraf

Husnain, Muhammad Shakir, Faiz Ali Yasir Shah Abbas and presented a document on

“Video Synthesis: Technique and Classification”. They present

techniques. The user wants to concentrate on the characteristics of the video.

Features such as color and movement. and voice, etc.

In 2016, Pawana Sharma1, Sachin Sharma2,

presented a report on “face detection and recognition with distance, and

SVM Hausdorff SURF”. They are used to obtain a better average error rate,

matching time, and accuracy result. Research work is limited to the acquisition

of facial recognition from a single image. The work can be extended into

several images at the same time. You can consider more and different parameters

in the future. In addition, new algorithms can be applied to improve face

detection and minimize execution time.

In 2009, Philippe Dreuw, Pascal

Steingrube, Harald Hanselmann and Hermann Ney presented a paper on the theme

“SURF-Face: facial recognition at the point sight restrictions

restrictions”, studied the use of SURF descriptors compared to the SIFT

descriptors for recognition Facial We have shown that using localized

aspiration features grid based approach rather than a based point detection

extraction of interest, the SURF descriptors and SIFT descriptors can be used

for face recognition, especially in combination with a grid-based compatibility

of coherence of the point of view.

Face Recognition

Algorithms

In this document,

the most environmentally friendly way to gain popularity for invariants is to

use the SIFT SURF and PCA strategies. The use of the SIFT and SURF algorithm

for detection capabilities and then follows PCA to fit in sentences of

rotation, expression and pose. Model recognition uses verification and

identification of two components 4.

PCA: these methods

are used for Eigen faces where images are small and reduce the size of data.

Image Compression Provides the most effective low-dimensional structure of the

facial model and each face image is represented as a vector of weighted sum

characteristics of the Eigen faces that are stored in the 1-D array. A linear

technique widely used in tactics based on the main aspect for FR. These

approximate objectives to solve the problem of popularity within an

illustration space lower than the image area 4.

SIFT: the SIFT

descriptor is invariable on scale, rotation, transformation, noise and is

highly distinctive. The characteristics of SIFT are four main steps in

detection and representation; (1) find the end of the scale space; (2) position

and filtering of key points; (3) orientation assignment; (4) descriptor of the

key point 4.

SURF: SURF

extracts key points from data set images and edited images. This coincides with

the key points between the modified image and each image in the database. In

the descriptor SURF it is invariable with a scale and the rotation

characteristics in the plane. It has two stages (1) detector of points of

interest and (2) descriptors of points of interest. The first stage, identify

the point of interest in the image. The use of the jute matrix to find the

approximate bearing is the difference of the Gaussian filter (DOG) used in the

SIFTS and in the points of interest of an image. The second stage, the

descriptors are used to extract the feature vectors at each point of interest

only in SIFT. Normally, SURF uses 64 SURF dimensions to reduce the cost of time

for both function matching and computation. SURF has a three times better

performance than SIFT 4.

In this document,

the Propose job first starts reading the input image and is preprocessed into a

grayscale image. Therefore, the features will be extracted from that image

using the SIFT and SURF algorithms respectively. It will be a produced image

consisting of both functions using SIFT and SURF.PCA will be applied directly

to that image. The goal of PCA is to extract the important characteristics of

facial data to delineate it as a set of new orthogonal variables that are

called main components. Now the coincidence will take place between the input

image and the image in which PCA is applied with different expressions,

contrast and rotation for invariable faces 4.

Object Detect Using Surf

In this document, first read

an integral image to store the object’s set and image, which will detect

characteristic points using SURF to form images and find the basis of the

strongest characteristic point in the threshold value. The strongest image

function is the extraction function and the corresponding function pairs and

controls a sufficient number of function pairs. Apply the RANSAC algorithm to

eliminate the wrong combined features 2.

In this document, two

different methods to characterize the extraction 1. Extraction of expressions

based on points of interest and 2. Based on grids.

The robust characteristics of

acceleration (SURF) are an invariant function of scale and rotation in the

plane. Contains detector and descriptor of points of interest.

1.

Interest point detection:

A difference in SIFT that uses

DoG to detect points of interest, SURF uses the determinant of the approximate

Hess matrix as the detector base. To identify the point of interest, we detect

structures similar to points at points where the determinant is maximum. The

integral images are used in the approximation of the Hesse matrix, which

drastically reduces the calculation time.

2.

Interest point description

SURF used the sum of Haar

ripple responses to describe the characteristic of a point of interest. Wavelet

filters were used to calculate the responses in the x and y directions. To

extract the descriptor, the first step is to construct a square region centered

on the point of interest and oriented according to the orientation decided by

the method of selection of the orientation introduced in. The region is also

divided into smaller 4 × 4 square subregions. This preserves important spatial

information. For each subregion, calculate Haar ripple responses in equidistant

5 × 5 sample points. For simplicity, we call right the Haar ripple response in

the horizontal direction and the Haar wave response in the vertical direction.

To increase the strength of geometric deformations and position errors, the dx

and dy responses are first weighed with a Gaussian centered on the point of

interest.

3.

Fast index for matching

To speed up the coincidence

step, use the Laplacian sign for the point of interest. Only the pair of points

with the same sign is combined with the characteristics.

Face Recognition Using PCA

In

this article we present the biometric identification technology that identifies

people based on their facial features. The innovation uses a camera or a webcam

to ensure images or video sequences that contain human aspects, recognizes and

tracks the face in the image, then performs a face recognition. The facial

recognition system has four parts. (1) acquisition and detection of facial

images, (2) facial image preprocessing, (3) facial features extraction and (4)

facial matching and recognition 5.

First

read the original color image. Then the extraction of the face region of a size

of 128×100 pixels which applies RGB to the grayscale image 5.

In

PCA, the faces are represented as a linear combination of weighted eigenvectors

called Eigen faces. These eigenvectors are obtained from the covariance matrix

of a set of training images called the basic function. The number of Eigen

faces that would be obtained will be equal to the number of images in the

training set. Eigen faces exploit the similarity between the pixels between

images in a data set by means of their covariance matrix 5.

Conclusions

From

many years the research in face recognition is an exciting area to come and

will keep many researchers, scientists and engineers busy. So we are using the

most flexible and efficient method for face recognition is SURF features in

face recognition and gives the detailed comparisons with SIFTS features.

Experimental results show that the SURF features perform only slightly better

in recognition rate than SIFT, but there is an obvious improvement on matching

speed. Therefore, SURF features are proven to be suitable for face recognition.