Color is one of the important features that make possible the recognition of image by human. J. Yue et al. in 2011, state that the color is one of the most widely used low-level visual features and is invariant to image size and orientation [3]. According to B.S. Manjunath et al., color features are the most expressive of all the visual features and have been extensively used in the image retrieval system [4].

Swain and Ballard have proposed a color matching method based on color indexing in early 1990s. This method is also known as color histogram intersection method and it is based on matching of color histograms [2]. In 2011, G. H. Liu et al. state in his paper that, color histogram is invariant to orientation and scale and this makes it powerful in image classification. Subsequently, color histogram-based color descriptor has been broadly studied and generally utilized in CBIR systems for its simplicity and effectiveness [5].

Another important element in visual perception is “Texture”. Texture also can be used to separate regions of interest in an image. According to Savita et al. in 2015, texture is also one of the most used low level visual features that refer to innate surface properties of an object and their relationship to the surrounding environment and it also contains important information about the structural arrangement of surfaces and their relationship with the surroundings [6].

According to M. Kokare et al., texture can be defined as, “a region in image has a constant texture if a set of local statistics or other local properties of the picture are constant, gradually differing or approximately periodic” [7]. In 1998, Y. S. Kim and W. Y. Kim explored that the methodology for existing texture classification can be roughly divided into three categories which are Statistical method, Structural method, and Model-based method [8].

Some of the algorithm for CBIR based on this texture as feature has been proposed by several researchers. In 2000s, a rotational invariant texture features using Dual-Tree Complex Wavelet Transform (DTCWT) for CBIR has been proposed by P. R. Hill, D. R. Bull, and C. N. Canagarajah, [9]. In 2003, C. M. Pun has proposed an effective rotation-invariant polar-wavelet texture feature for image retrieval [10]. Other than that, in 2005 M. Kakore et al. proposed texture feature for image retrieval that supports shift invariant by using a set of two-dimensional Dual Tree Complex Wavelet Transform (DTCWT) and Dual Tree Rotated Complex Wavelet Filters (DTRCWF) jointly [11]. J. Han and K. K. Ma in 2007 has presented texture features for image retrieval using Gabor wavelet. In their proposal, mean and standard deviation values of transformed coefficients computed individually found to be rotation and scale invariant [12]. In 2009, a texture image retrieval methodology based on non-tensor product wavelet filter banks has been proposed by Z. He, X. You, and Y. Yuan [13].

In content-based image retrieval system, another important visual feature is shape. According to D. Zhang and G. Lu in 2004, shape is one of the basic features used to describe image content and it is also can be used to provide powerful information. Nevertheless, shape representation and description is a difficult task because when a 3-D real world object is projected onto a 2-D image plane, one dimension of object information is lost. As a result, the shape extracted from the image only partially represents the projected object. Further, shape is often corrupted with noise, defects, arbitrary distortion and occlusion [14].

According to Rao & Kumar in 2012, shape representation and description methods can be divided into two main categories which is boundary-based and region-based [2]. Savita Gandhani et al. in 2015 stated that the boundary-based technique will use only the contour or border of the object and completely ignores its interior, whereas region-based technique apply segmentation to divide an image into different regions/segments, by setting threshold values according to the desirable results [6].

According to Sharmin Siddique, the most successful representations for shape categories are Fourier Descriptor and Moment Invariants. She stated in her paper, the main idea of Fourier Descriptor is to use the Fourier transformed boundary as the shape feature, whereas Moment invariants is to use region-based moments, which are invariant to transformations as the shape feature [15].

In content-based image retrieval system, some key parameters have been defining and evaluate based on precision, recall and the response time. The image retrieval precision rate and recall rate can be defined as this following formula:

p r e c i s i o n = R N r e c a l l = R M

The N is stand for the number of images returned for the query, whereas R is images associated with the example in the result. Besides, M is stand for images associated with the example in test set S (saturation) [16].

Based on comparison study by D. Guoyong et al. in 2011, for the texture feature technique, the precision percentage is 60.1%, recall percentage score is 72.3%, and response time is 1372 millisecond. Other than that, for the color feature technique, the precision percentage score is 53.7%, recall percentage is 65.2%, and response time is 824 milliseconds. Besides that, for the shape feature technique, the precision percentage is 62.2%, recall percentage is 70.8%, and response time is 2544 millisecond. Last but not least, for the combined feature technique, the precision percentage is 79.6%, recall percentage is 88.3%, and response time is 3861 milliseconds [16] (Figure 2).

Based on the comparison study, for the first parameter which is precision percentage, combined feature is the highest. For second parameter, which is recall percentage, combined feature is the highest. Besides, for the third parameter which is response time, color feature is the fastest. Based on that comparison study, the combination or hybrid features technique which is combination of texture, color and shape strongly improved precision and recall, although it has longer response time.