Special attention has been given to the lack of geometrical information encoded

Special attention has been given to the lack of

This preview shows page 74 - 76 out of 94 pages.

Special attention has been given to the lack of geometrical information encoded by the traditional BoW representation [5,13]. The spatial arrange- ment of visual words in images is important to understand image semantics and is often crucial to distinguish different classes of scenes or objects. In that direction, approaches are proposed for image classification [10,9] and re- trieval [8,11,12,13]. In the classification scenario, usually relied on Support Vector Machines (SVM), the high dimensionality of vectors does not degrade effectiveness, because SVMs suffer less from the curse of the dimensionality. In this paper we present an image retrieval and classification method based on Bag of Visual Words (BofVW) through object recognition, by using feature vectors computed on the image. 2 Background The most popular and effective approach to represent visual content nowa- days is based on visual dictionaries, which generate the BoW representation [1]. One of the benefits of using such a representation is its ability to encode local properties into a single feature vector per image. To generate a BoW rep- resentation, one must first create the visual dictionary. Image local features, usually obtained by SIFT descriptor [4] computed on the detected points of interest [2], or in a dense grid [3], are clustered in the feature space. Each cluster represents a visual word and tends to contain patches with similar appearance. Several recent works in the area present interesting advances for creating better dictionaries and better coding techniques. However, many of the recent advances built over the visual dictionary model relies on encoding geometrical information of visual words. Researchers faced the problem of having many different images with identical or very similar color histograms, motivating the creation of new methods for encoding the spatial arrangement of colors,
Image of page 74
Image Retrieval and Classification Through Object Recognition 69 like, for example by using color correlograms or color-coherence vectors [5]. This issue is being revisited nowadays with the visual dictionary model, for discriminating image content and encoding image semantics. Consider the images shown in Fig.1. They have different semantics but their BoW repre- sentations are very similar. Fig. 1: Examples of images with different semantics but similar BoW. This figure is partially a reprint from [5]. Graph-based approachs have been used to represent object relations within images. The main motivation for that relies on their invariance to transfor- mations like rotation and translation. From an image collection, firstly, it is necessary to detect all the interest points and, then, cluster the descriptors of the interest points in the feature space. Afterwards, a set of connected graphs is defined in order to generate the visual-word dictionary from the prototypes of the clusters. Using this dictionary we can represent the image, and encode the spatial relationships of visual words. The process to generate the Bags
Image of page 75
Image of page 76

You've reached the end of your free preview.

Want to read all 94 pages?

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Stuck? We have tutors online 24/7 who can help you get unstuck.
A+ icon
Ask Expert Tutors You can ask You can ask You can ask (will expire )
Answers in as fast as 15 minutes