978-1-5386-5995-3/18/$31.00 ©2018 IEEE Narendra Kumar Rout 1 Department of Computer Application National Institute of Technology Raipur CG, India narendrarout@gmail.com Mitul Kumar Ahirwal 2 Department of Computer Application National Institute of Technology Raipur CG, India ahirwalmitul@gmail.com I. INTRODUCTION With the expanded use of smart phones, internet, and digital cameras, the amount of images generated has increased [1] as well as its applications in classification and detection has exponentially increased such as crime prevention [2], security check [3], remote sensing, geography [4], medical diagnosis [5], architecture [6], advertising, design, fashion, and publishing [4]. As image database consists of different type/categories of images, content based image retrieval (CBIR) is a method of finding similar images. An image represents hundreds and thousands of words, which is indexed by corresponding visual elements, like color, shape, and texture in CBIR. This signifies that the relevant images are retrieved from an image database with the help of derived image features as automatic consequences. It also boosts the intermediary interface between computer system and the user. In recent year, images are getting important position for accessing visual representation. Munesawang et al. 2004 [7] mentioned in their study that besides text retrieval, image retrieval could be the most effective way of searching. For making image retrieval possible, CBIR takes the power of accessing the image. Now-a-days this system can be used for commercial purpose, for example IBM’s Query By Image Content (QBIC) [8], VisualSEEk [9], NeTra [10], WebSEEk [11]. The purpose of CBIR is to study the details of an image by low level features (e.g., colour, texture, shape etc.) and create its correspondent image index by organizing feature vectors [6]. In the past, many authors have defined various features like color, shape, texture of an image. Color is an effective feature for perception of an image containing global features [6] and hugely characterizes visual features that make the CBIR system most robust [12]. It is reported that the color is implemented in image retrieval system easily, since it sharpens its visual features of an image, regardless of image size and orientation [12]. But the drawback behind the color is its restriction on individual matter of opinion and resolution [13]. Mainly color image consists of three basic color components, i.e., red, green and blue. There are many color features widely used in image retrieval [14]. These are: color moments [15], color histogram [16], color correlogram [14], color coherence vector [17], and color difference histogram [18]. Color is easy to implement and needs few storage requirements as requires simple computation [6] [19]. Another functional and subjective low level input feature is its texture which is very effective [20]. Texture is a unit of visual form of consistency properties of a surface having no connection with any single color and intensity [21]. Wang et al. 2014 [13] recommended that the texture is a visual feature deliberated to innate the coarseness and repetitive phenomenon of surfaces inside the image [13]. Texture has many features in image processing like Gray-level co-occurrence matrix (GLCM) [13], Markov random field (MRF) model, edge histogram descriptor (EHD), Local binary pattern (LBP) etc. [22]. The GLCM provides the information on different directions with A Content Based Image Retrieval System: Analysis of Individual and Mixed Image Features Abstract— Inspite of progress in computer vision, systems still find it very challenging to identify and choose images that are similar to a given one without a thorough process of training. However computers are good at finding individual features within a given image. Hence an effort has been made through this paper to explore the power of individual feature for different images. Though this approach is simple but there haven’t been attempted to thoroughly study its utility towards computer based image recognition. As the first step to that a limited set of color, texture and shape features have been chosen for the purpose of this study. The feature computation was done on each of these images using their standard mathematical formula and similarity is computed based on Euclidean distance between the values. Similarity detection performance was calculated using precision and recall. After the experiment, the precision rate of tested query images results better uses of all features with equal weight for different images. This content based image retrieval system encourages the identification of methods using feature isolation and optimization of associating weight to each of the features for different type of images. Keywords— Content based image retrieval; color feature; texture feature; shape feature. International Conference on Recent Innovations in Electrical, Electronics & Communication Engineering - (ICRIEECE) 2561 Authorized licensed use limited to: MAULANA AZAD NATIONAL INSTITUTE OF TECHNOLOGY. Downloaded on May 22,2021 at 14:52:47 UTC from IEEE Xplore. Restrictions apply.