Similar Image Retrieval Using Color Histogram in HSV Space and SIFT Descriptor with FLANN
This work proposes an efficient method for similar image retrieval based on color histogram and local feature descriptors. It has proven that the SIFT descriptor achieves the best performance among all the local descriptors. However, the high dimensionali
- PDF / 1,004,521 Bytes
- 9 Pages / 439.37 x 666.142 pts Page_size
- 45 Downloads / 293 Views
Abstract This work proposes an efficient method for similar image retrieval based on color histogram and local feature descriptors. It has proven that the SIFT descriptor achieves the best performance among all the local descriptors. However, the high dimensionality of the scale-invariant feature transformation (SIFT) feature descriptor brought difficulties for image matching. We first extract color histogram in HSV and discard the image whose histogram is less than the threshold to the query image. Then, for the remaining images, we extract SIFT feature and use Fast Library for Approximate Nearest Neighbors (FLANN) matching algorithm to get a score for every candidate image. Finally, and most important conception proposed by us, we weight both color histogram and SIFT feature to get a score rank as our retrieval result. From the procedure, we have advantages in both time consumption and result accuracy. Experimental results demonstrate that the performance of this scheme is efficient.
Keywords SIFT Color histogram in HSV space retrieval Image matching
FLANN Similar image
1 Introduction With the rapid development of network technology and digital technology, there are many similar images photographed from different viewpoints but with the same scene or object. Content-based image retrieval extracts feature information in Y. Wang (&) H. Li L. Wang School of Information Science and Engineering, Lanzhou University, Lanzhou 730000, China e-mail: [email protected] H. Li e-mail: [email protected]
Z. Wen and T. Li (eds.), Foundations of Intelligent Systems, Advances in Intelligent Systems and Computing 277, DOI: 10.1007/978-3-642-54924-3_103, Springer-Verlag Berlin Heidelberg 2014
1085
1086
Y. Wang et al.
pixels of different images as a clue to calculate the similarity between images. CBIR can work well to find similar images in a dataset of a query image. Traditional CBIR methods in feature detection and matching revolve around low-level features such as color, shape, and texture. Color histogram depicts the ratio of different colors occupied in the whole image. It takes no account of the spatial position of every color and cannot depict object in the image. However, it can be used in combination with local feature descriptors to get considerable result. The scale-invariant feature transformation (SIFT) algorithm, proposed in [1] by Lowe in 2004, is found highly distinctive, and invariant to scale, rotation, and illumination changes. According to the evaluation [2], the SIFT descriptor [1] achieves the best performance among all the local descriptors. Because one of the drawback of SIFT is the computational complexity of the algorithm increases rapidly with the number of keypoints, especially at the matching step due to the high dimensionality of the SIFT feature descriptor, many approximate nearest neighbor search algorithm have been proposed, such as kd-tree [3] and vocabulary tree [4]. In [5], Muja and Lowe compared many different algorithms for approximate nearest neighbor search on da
Data Loading...
