Flower Image Retrieval Using LBP, Wavelet Moments, Gabor Wavelet Features and Multiple Distance Functions

Abstract

Flower image retrieval is a significant and challenging problem in content-based image retrieval. In this paper, we propose a content-based method for retrieving flower images of specified specie from a database of flower images of various species. Firstly, we use wavelet moment, Gabor wavelet and Local Binary Pattern (LBP) independently to characterize all flower images in the database. Secondly, we represent a query flower image with Gabor wavelet, wavelet moment, and LBP individually and search images in the database analogous to the query image. The retrieval is accomplished through calculating similarities between the query image and the database images by employing a set of distance functions. Experimental evaluation of the approach reveals that the Gabor wavelet achieves superiority over the wavelet moment and LBP considerably. It is also indicated that the retrieval outcome can be improved through concatenating the Gabor wavelet, wavelet moment, and LBP features rather than utilizing them individually. 1 Introduction At present world, with the availability of image capturing devices such as digital cameras, mobile cameras, webcams, video cameras, image scanners, the size of digital image collection is increasing at a great rate. A computer system for image searching, browsing and retrieving from a large database of digital images is called an image retrieval system. It is a tool, required by users from various domains, including remote sensing, medicine, crime prevention, architecture, publishing, fashion, etc. From this point of view, there have been developed many prevalent image retrieval systems. There are two frameworks of image retrieval such as text-based image retrieval and content-based Image Retrieval. The text-based approach started since the early 1960s. In text-based image retrieval system, for finding similar types of images it is needed to search by the accurate textual description, filenames, caption, keywords etc. Because the images are annotated in the database by textual description, filenames, caption, keywords etc. But text-based image retrieval is oldest image searching system. There are many several problems of text-based image retrieval methods. Firstly, annotation is never complete because it depends on the goal of the annotation, some mistakes like spelling error, spelling difference (US vs.UK), weird abbreviation (particularly medical) etc. The use of keywords is not only complex but also insufficient to represent the content by the size of image databases growing etc. The keywords are not unique for one kind of searching. The problems of image queries by text-based approaches cannot be described at all, but tap into the visual features of images .For example, a query for all the images in the database with "classroom" in it will good results if we annotate all the images containing classroom but for the same annotations, a specific search for images with a chair, table, computer or student in tit will fail. Another problem with text based approach is the lack of uniform