International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056 Volume: 03 Issue: 05 | May-2016 www.irjet.net p-ISSN: 2395-0072 © 2016, IRJET | Impact Factor value: 4.45 | ISO 9001:2008 Certified Journal | Page 1497 Spatial pooling of heterogeneous features for image classification using GPP Charulata Rathi 1 , S.V.Jain 2 Student, Computer Science and Eng., ShriRamdeobaba College of Engineering and Management, Nagpur, India 1 Assistant Professor, Computer Science and Eng., ShriRamdeobaba College of Eng. and Management, Nagpur, India 2 Abstract—Geometric phrase pooling (GPP) refers to extraction of spatial features of images in both x and y coordinate. It holds an important stake in the process of image classification. Application of GPP for image extraction is done using BOF model.The Bag-of- Features (BoF) is a widespread model that targets to epitomize images as a loose collection of features devoid of the use of some spatial statistics. As a result of its straightforwardness and decent enactment its attractiveness have significantly increased for image classification. BoF has progressed from method named as texton used in texture analysis. For execution of the BoF model four selections are necessary to be done, these selections are method to be used for sampling patches, procedures used to define them, characterization of the obtained spatial dissemination and finally classification of images built on the outcome. With the use of various methods such as Region of Interest (ROI) and extraction of multiple descriptors, disadvantages of BoF model can be neutralised. In spite of using all the above mentioned techniques rational assimilation of all the modules is not capable of resolving the disadvantages. And to overcome the above mentioned issue, this paper suggest an excellent framework with spatial assembling of heterogeneous features .BOF model is appliedby means ofsucceedingprocedures , speeded up robust feature (SURF) descriptor and canny edge map detector are used for extraction of spatial heterogeneous features, then feature pooling is used for construction of codebook, Support vector machine (SVM)is used for image classification and lastly color, texture and shape based techniques are used for image retrieval. Keywords— Image classification, BoF Model, K-medoid clustering, Image matching, SURF. I. INTRODUCTION Image classification has been one of the most prominent procedure in computer vision. It comprises of image processing and image analysis ,this technique primarily aims to convert input 2D image to another by performing from pixel wise operation such as enhancement in the contrast level , local or global descriptors extraction, noise removal or edge extraction. Some of the main applications of image classifications are scene matching, remote sensing, object detection, medical applications, Google goggle and many more.One of the prominent algorithm applied for image classification is Bag-Of-Feature, which recommends enhanced illustration of images by statistics-based model [1,2,3].To achieve this speeded up robust feature(SURF)is used. using SURF, descriptors from the input image are extracted by selecting some prominent features based on pixels depicting Swift changes in intensity values in both the planes. In SURF a histogram is constructed along the local neighbourhood of each key point by construction of a descriptor vector. [4,5]This paper proposes a technique for image retrieval using SURF, prominent features extraction is done. Robustness and reduction in run time are some of the primary advantages that SURF depicts over SIFT. For edge map extraction canny edge map detector is used. one of the many advantages of canny is its preciseness. Procedure involve in combining of SURF descriptor and canny edge map is called as feature pooling, which is used for codebook construction. SVM classifier is trained using training dataset, so as to label the images into restricted categories.[6]feature based extraction technique such as texture of the object, color of the image and shape of the object are used for retrieval of images. II. RELATED WORK In the year 2014, LingxiXie, Qi Tian, Senior Member, IEEE, Meng Wang, and Bo Zhang proposed Spatial Pooling of Heterogeneous Features for Image Classification[7]. In this paper texture and edge based local features of input image are extracted using SIFT (Scale Invariant feature Transform) descriptor, then midlevel image representation upon complementary features is build using geometric visual phrase and finally spatial weighting of the image is calculated using the smoothed edge map to capture the image saliency. Construction of codebook for database vocabulary is done using K-means algorithm. Although BoF model is successful using SIFT, some of the disadvantages of SIFT are it suffers from Synonymy and polysemy[7,8,9], time complexity is more as compared to SURF, also process complexity is