Journal for Research | Volume 02 | Issue 06 | August 2016

ISSN: 2395-7549

Object Detection and Recognition: A survey

Anjana Mittal Dr. Sidhhartha Choubey
M.E. Scholar Associate Professor
Department of Computer Science & Engineering Department of Computer Science & Engineering
SSTC Bhilai (C.G.) SSTC Bhilai (C.G.)

Abstract
The efficiency and quality of a feature descriptor are critical to the user experience of many computer vision applications.
However, the existing descriptors are either too computationally expensive to achieve real-time performance, or not sufficiently
distinctive to identify correct matches from a large database with various transformations. In this paper, we propose a highly
efficient and distinctive binary descriptor, called local difference binary (LDB). LDB directly computes a binary string for an
image patch using simple intensity and gradient difference tests on pair wise grid cells within the patch. A multiple-gridding
strategy and a salient bit-selection method are applied to capture the distinct patterns of the patch at different spatial granularities.
Experimental results demonstrate that compared to the existing state-of-the-art binary descriptors, primarily designed for speed,
LDB has similar construction efficiency, while achieving a greater accuracy and faster speed for mobile object recognition and
tracking tasks.
Keywords: Object recognition, binary descriptor, spatial granularity, local difference binary and multiple - gridding
_______________________________________________________________________________________________________

I. INTRODUCTION

The object recognition is a challenging task in the real world [1]. The problem is that each object makes a different perspective to
the human mind with the position of the object, pose and background. A human brain can effortlessly by its own hypothesis and
biological inspirations. A method to solve the problem is to create a large database of the similar images with the variations is
done. The different methods have developed over the years for the recognition of the object from the given set of images.
Some of the basic components for the object recognition are:
1) Base Model: The model describes the functional descriptions to the feature information. The functional information models
for the size, shape and color of the features. Certain schemes are applied for facilitating the candidate objects for
consideration.
2) Feature detector: A feature detector detects the objects locations forming the hypotheses. The recognition depends on the
organization of the database the category of objet to be recognized.
3) Hypothesis formation and verification: A verifier or classifier can be considered for finding the object with the maximum
relevance refining the likelihood of the correct object.
The object recognition task is a challenging issue due to the following factors:
1) Scene Constancy: The various conditions such as illumination, background, camera parameters, etc affect the image of the
object dramatically. In such a case, the recognition may variated in different conditions for the same image.
2) Dimensionality: The image or video obtained can have three dimensional objects which are represented into two
dimensions. Here the features detected in two dimensional image space which are originally three dimensional features may
appear differently and can cause complexity in representations.
3) Number of objects: The feature detection for the objects in an image or video sequence is easier when the number of
objects is less. Selecting the appropriate features for object recognition is directly proportional to the number of objects in a
scene. Therefore, increasing the number of objects increases the effort of feature selection.
4) Occlusion: occlusion is a serious problem in an image processing task as it results in the absence of expected features and
generation of unexpected features. With the increase in number of objects in an image, the probability of occlusion also
increases.
The field of image processing and analysis is very broad, encompassing a wide variety of research issues, from efficient
encoding of images and video sequences, through image enhancement and restoration to image segmentation, recovering spatial
shape from shading and pattern distortions, learning about 3D from multiple unregistered 2D images, of image sequences, and
high level image understanding topics.

II. LITERATURE SURVEY

In 2012 Reza Oji [2], proposed a method for object recognition with boundary detection with the ASHIFT (Affine Scale
Invariant Feature Transform) and region merging. Initially the segmentation is done o partition the image into smaller portions.
The images are simulated with two camera orientations. The key point descriptor relative to the orientation can get invariance in
rotation. The histogram is created of local gradients to select the scale. The highest peak shows the direction of the local gradient.

All rights reserved by www.journalforresearch.org 28

 Object Detection and Recognition: A survey
(J4R/ Volume 02 / Issue 06 / 006)

The regions in the image that corresponds to the object are identified with the Euclidean measure. The lower the distance hence
more is the similarity. The merging process is done after the ASHIFT is performed. The basic aim of merging is to merge the
non-object regions so that the object gets highlighted.
In 2012, V. Subbaroyan, Dr. Selvakumar Raja [3], proposed an approach for the recognition of the multiple objects by
considering the features from spatial as well as frequency domain. The input image is converted into gray scale image. The Sobel
operator is used to find the gradient of the image intensity. The color features of the image are obtained with the help of the
global color histogram. The two level wavelet transform is done to obtain the feature with higher frequency. The spatial features
are obtained from three colo moments for the basic colors. Both features are fused to obtain the recognized object.
Michael Calonder et al [5], proposed a method where the image patches are classified with respect to the pair wise intensity
values. The Nave Bayesian Classifier recognizes the patches from the different viewpoints. A clear advantage of binary
descriptors is that the Hamming distance. The individual bits are compared with the intensities of pairs of points. The positions
of the pixels are pre-selected randomly according to a Gaussian distribution around the patch center. The kernels are used to
smooth the patches before intensity difference and spatial pair of images.
Chiverton et al. [6] proposed a new fully automatic object tracking and segmentation framework. The framework consists of a
motion-based bootstrapping algorithm concurrent to a shape-based active contour. They propose two approaches of increasing
computational intensity and accuracy that statistically estimate the foreground from the potential mixture of foreground and
background enveloped by an alpha hull.
Zhenjun et al. [7] used combined feature set which is built using color histogram (HC) bins and gradient orientation histogram
(HOG) bins considering the color and contour representation of an object for object detection. The combined feature set is the
evolvement of color, edge orientation histograms and SIFT descriptors.
R. Saravanakumar et al. [8] represented the objects using the properties of the HSV color space. Adaptive k-means clustering
algorithm was applied to cluster objects centroids color values and co-ordinates were sent to next frame for clustering. A three
dimensional representation of the HSV color space is a hexacone, with the central vertical axis representing intensity.
Xiaobai Liu et al. [9] proposed hybrid online templates for object detection which uses different features such as flatness,
texture, or edge/corner. The template consists of multiple types of features, including sketches/edges, texture regions, and
flatness regions. Sketch/edge regions usually consist of various links, ridges, such as corners and junctions. Texture regions are a
large number of objects that are either too small or too distant to the camera. In contrast, flatness regions are always filled with
homogeneous color or intensity. The limitation of this method is; as the discriminative power of features change along with the
object movements, the hybrid template should be adaptively updated by either adjusting the feature confidences, or substituting
the old features with the newly discovered ones from the currently observed frames.
Liu et al. [10] presented a novel semiautomatic segmentation method for single video object extraction. Proposed method
formulates the separation of the video objects from the background as a classification problem. Each frame was divided into
small blocks of uniform size, which are called object blocks if the centering pixels belong to the object, or background blocks
otherwise. After a manual segmentation of the first frame, the blocks of this frame were used as the training samples for the
object-background classifier.
Johnsen et al. [11] used Approximated median filter to perform background modeling. For the implementation, better results
were obtained by scaling the increment and decrement by a step factor if the absolute difference between the current pixel and
the median-modeled background pixel is bigger than a threshold.
Victoria Yanulevskaya et al [12] proposed an object-based visual attention theory for the task of salient object detection. They
assume proto-object being a unit of attention and argue that notion of an object should be taken into account while assessing
object saliency.
Carlos Cuevas and Narciso Garca [13], proposed the background modeling algorithm to detect real time moving object. The
proposed method combined the background model and foreground model to detect an object from complex image very quality
taken by non-completely static cameras. It finds the bandwidth matrices for the kernels which are used in background modeling.
This proposed method updating the background model for reducing the misdetections. Bahadir Karasulu and Serdar Korukoglu
[13], proposed Moving object detection and tracking by using annealed background subtraction method in videos. Current frame
is subtracted from image and it is used to classify the pixel either foreground or background by comparing the difference with the
threshold. Simulated annealing (SA) technique is used to rectify the p median problem. The total weighted distance between
demand points (nodes) and the closest facilities to demand points are used to find the p number. SA-based hybrid method is
developed for performance optimization of back, which is used to detect and track object(s) in videos.
Ling CAI Et Al [14], proposed a stereo vision-based model for multi-object detection and tracking in surveillance.
Illumination variation, shadow interference, and object occlusion problems are overcome by using stereo model. They identified
the feature points after they projected into 2D ground plane. For grouping the projected points according to their height values
and locations on the plane, a kernel-based clustering algorithm is used.
Bangjun Lei and Li-Qun Xu [15], proposed a detection and tracking of objects in wide range of outdoor surveillance and
monitoring scenarios in real-time video analysis system. Adaptive background modeling technique is used to extract the
foreground regions. A blob analysis is used for object tracking. It gives better result for non-crowded and the static state of scene.

All rights reserved by www.journalforresearch.org 29

 Object Detection and Recognition: A survey
(J4R/ Volume 02 / Issue 06 / 006)

Feature Descriptor:
Feature point descriptors are widely used in many computer visions tasks such as marker less-augmented reality simultaneous
localization and mapping. Some available descriptors are:-
The SIFT descriptor, proposed over a decade ago, has been widely adopted as one of the highest quality options. However, it
imposes a heavy computation burden. This drawback has drawn extensive efforts for optimizing its speed without compromising
too much quality.
SURF is arguably the most noticeable. But recent experiments have shown that the SURF descriptor is still too
computationally heavy. Thus only a limited number of points can be handled for realtime applications such as AR, especially
for handheld devices such as smart phones and tablets.
BRISK descriptor is a method for key point detection, description and matching. A novel scale-space FAST-based detector
reveals the descriptor sped which is combined with the assembly of a bit-string descriptor whose intensity is compared with each
neighborhood key point by sampling dedication.
BRIEF descriptor is proposed as an efficient feature point descriptor which uses binary strings. BRIEF using simple intensity
difference tests for computation and even when using relative few bits, it is highly discriminative. The descriptor is much faster
than other state-of-the-art ones, not only in matching and construction but also yields higher rates of recognition.
ORB descriptor is proposed on the basis of very fast binary BRIEF descriptor, which is resistant to noise and invariant to
Rotation. SIFT is at two orders of magnitude slower than ORB, it is an efficient alternative to SIFT or SURF. It is an efficient
alternative to SIFT or SURF. Its efficiency is tested on smart phone while patch-tracking and objects detection including several
real-world application.
FREAK represents retina-inspired and coarse-to-fine descriptor to improve the performance of current image descriptors. It
performs better than other state-of-the-art key point descriptor while remaining faster to compute with lower memory load and
also more robust than SIFT, SURF or BRISK.
Lightweight binary descriptors such as BRISK, FREAK, BRIEF and its variant rBRIEF (or ORB descriptor), become
increasingly popular as they are very efficient to store and to match. Raw intensities of a subset of pixels within an image patch
for binary tests, and thus have low discriminative ability. Lack of distinctiveness incurs an enormous number of false matches
when matching against a large database. The basic problem with existing descriptors is high computing complexity, matching
and storing the feature point descriptors.

III. PROPOSED METHODOLOGY

The method is a two-step process:

1) Smoothing: The image is smoothened at the corner for the boundary values of the image. It obtains the homogeneous points
and corner points of the image and selects the feature vectors.
2) Object Recognition: The process of object recognition is done from the set of features obtained after smoothing of the
image. Local Difference Binary (LDB) is done for the object recognition. The image is divided into small blocks and the
information from each pixel is extracted. This information describes the efficiency of the process and is the average of the
pixel intensity values. The first-order gradient is obtained from the average intensities that confine the variation in the
images. The level of block division needs to be fine as more granularity leads to more distinct recognition. For flow of
proposed methodology refer Fig 1.

All rights reserved by www.journalforresearch.org 30

 Object Detection and Recognition: A survey
(J4R/ Volume 02 / Issue 06 / 006)

Input the image

Perform the smoothing

and obtain the boundaries

Divide the image into n

blocks

Find the average intensity

of the pixel valued

Perform first order

gradient

Perform recognition
through matching

Output the recognized

object
Fig. 1: Flow Chart of proposed methodology.

IV. CONCLUSION

In this paper, I only test some Images which were taken from the standard Pascal Voc dataset and first I recognize the object with
the existing algorithm means bounding box method, in bounding box method they recognize the whole object sometimes I did
not need of whole object. In our proposed work I did not work on bounding box method, first of all I will use gridding method
and grid it in 3*3 gridding and then I use edge detection method to find out the edges of the particular object only that is my
main recognition. The proposed method performs the recognition with the boundary of the extraction of the object with the
gridding for better recognition. After that report I will test our algorithm in some more Images and also compare our work with
the existing work also.

REFERENCES
[1] Xin Yang, Kwang-Ting (Tim) Cheng,Local Difference Binary for Ultrafast and Distinctive Feature Description", IEEE Transactions on Pattern Analysis
and Machine Intelligence", VOL- 36, NO-1, JANUARY-2014.
[2] Reza Oji, An Automatic Algorithm for Object Recognition and Detection Based on ASHIFT Key points, Signal & Image Processing: An International
Journal (SIPIJ) Vol.3, No.5, pp. 29-39, 2012.
[3] V. Subbaroyan, Dr. Selvakumar Raja, Novel Feature Fusion Method of Object Recognition Using Wavelet Transform, International Journal of Scientific &
Engineering Research, Vol 3, Issue 9, 2012.10.
[4] ShulinYang, LiefengBo, Jue Wang and Linda Shapiro, Unsupervised Template Learning for Fine-Grained Object Recognition, Advances in Neural
Information Processing Systems, pp. 31223130, 2012.
[5] Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua, BRIEF: Binary Robust Independent Elementary Features, ECCV10 Proceedings of
the 11th European conference on Computer vision: Part IV, pp. 778-792, 2010.
[6] John Chiverton, Xianghua Xie and Majid Mirmehdi,Automatic Bootstrapping and Tracking of Object Contours , IEEE Transactions on Image
Processing, Vol. 21, No. 3, March 2012.
[7] Zhenjun Han, Qixiang Ye, Jianbin Jiao, Online feature evaluation for object tracking Using kalman filter 978- 1-4244-2175-6/08/$25.00 2008 IEEE.

All rights reserved by www.journalforresearch.org 31

 Object Detection and Recognition: A survey
(J4R/ Volume 02 / Issue 06 / 006)

[8] S. Saravanakumar, A. Vadivel and C.G. Saneem Ahmed, Human object tracking in video sequences ICTACT Journal on Image and Video Processing,
August 2011, Volume: 02, Issue: 01.
[9] Xiaobai Liu, Liang Lin, Shuicheng Yan, Hai Jin, and Wenbin Jiang, Adaptive Object Tracking by Learning Hybrid Template Online ,IEEE Transactions
On Circuits And Systems For Video Technology, Vol. 21, No. 11, November 2011.
[10] Yi Liu, Yuan F. Zheng, Video Object Segmentation and Tracking Using -Learning Classification , IEEE transactions on circuits and systems for video
technology, vol. 15, no. 7, July 2005.
[11] Swantje Johnsen and Ashley Tews, Real-Time Object Tracking and Classification Using a Static Camera Proceedings of the IEEE ICRA 2009 Workshop
on People Detection and Tracking Kobe, Japan, May 2009.
[12] Victoria Yanulevskaya, Jasper Uijlings, Jan-Mark Geusebroek, Salient object detection: From pixels to segments, Image and Vision Computing Vol. 31,
Pg. No. 3142, 2013.
[13] Carlos Cuevas and Narciso Garca , Improved background modeling for real-time spatio-temporal nonparametric moving object detection strategies,
Image and Vision Computing Vol.31, Pg. No. 616630,2013.
[14] Ling CAI et. al., Multi-object detection and tracking by stereo vision, Pattern Recognition Vol.43, Pg. No. 40284041, 2010.
[15] Bangjun Lei and Li-Qun Xu, Real-time outdoor video surveillance with robust foreground extraction and object tracking via multi-state transition
management, Pattern Recognition Letters Vol.27, Pg. No. 1816 1825, 2006.

All rights reserved by www.journalforresearch.org 32