Cuckoo Search Algorithm Based Feature Selection in Image Retrieval System

Efficiency an d retrieval time are very important issues in any content-based image retrieval system. In this study, an efficient image retrieval system was introduced depending on several features extracted from the database images, namely color moment (Mean, Standard Deviation), GLCM, and DWT( only LL-sub band). To increase the retrieval speed, Cuckoo search algorithm was used to select the important positions that contain full power features from the (LL-sub band). On using the Cuckoo search algorithm, only (50) important positions were chosen out of the total (24576) positions within (LL- sub band). These positions were stored for later use when entering a query image. Thus, the time taken to retrieve images was greatly reduced and this process also increased the efficiency of the system due to the fact that the selected positions gave the lowest distance measures between the query images and the similar images when evaluated using Manhattan distance measure. Two effectual performance measures (precision & recall) were used to calculate the accuracy of the system. The findings proved the system efficiency when compared to other previous works.

The retrieval system in this research subsumes the following steps:  Features extraction using two techniques: color and texture.  Features locations selection using Cuckoo search algorithm.  Similarity measure.

Features Extraction
Feature extraction forms a crucial step in any CBIR system. The increase in data size, represented by the number of image pixels, increases the size of the search space, which in turn makes it difficult to perform additional data processing. As such, an appropriate representation of image pixels is required. Two kinds of features are used to represent image: A Low-level feature which includes color, texture, and shape features and A High-level feature which is extracted by using machine learning techniques (Atif et al. 2018) (Mohammad & Zahra 2017 Where hij is i th color channel & j th pixel . B. Texture Feature Extraction Texture is an important feature of the image retrieval system. It is also a low-level visual feature used to distinguish images with high efficiency (Lakshmi et al. 2014).  GLCM One of the most prominent statistical methods used to derive texture feature is the Gray level co-occurrence matrix of which four characteristics can be identified by using the following equations (K. P. Rajeswari et al. 2018) (Lakshmi et al. 2014): Where i is no. of rows , j is no. of columns of the image x .  Wavelet Transform Wavelet forms a useful tool to analyze signals and images. It is used to extract features from an image by decomposing it into the following 4-sub images, each of which is ¼ size of the original image: (

2.2.Feature Selection Using Cuckoo Search Algorithm.
In the feature selection process, a subset of the original features is selected. This results in a reduction of the data, speed and efficiency in data mining (Lakshmi et al. 2014). Here Cuckoo search is used to select the features. CSA is one of the most recent meta-heuristic algorithms inspired by nature. It was developed by Yang and Deb in (2009). What distinguishes cuckoo birds as well as their beautiful sounds is the strategy they possess for aggressive reproduction where some kinds of them put eggs in host nests and remove the host's eggs so as to increase the likelihood of hatching their eggs.
The following three rules are used to simplify the description of cuckoo search strategy:  Each bird puts one egg at a time, and casts its egg in the nest that has been chosen at random.  The best nests that contain high quality eggs will move on to the next generation. Were, N represent no. of features in each vector, vquery (i) is query image feature vector , and vdbase (i) is database image feature vector

The Proposed Image Retrieval System
To understand the overall steps in the proposed image retrieval system see figure 1: Figure 1. The Proposed System Step1: Load all images database; each of size (256×384) Step 2: Compute the Mean, Standard Deviation, and GLCM of all images and store them (18 features for each image).

Materials
This work was performed using Matlab R2017b language. A collection of Coral database images was also taken represented by 4 classes (Bus, Food, Elephant and Dinosaur), with 25 images taken from each class. Table (1) illustrates the parameter values used in Cuckoo search: Table (

Methods of Evaluation
To evaluate the accuracy of the proposed image retrieval system, two effectual performance measures have been used, namely precision & recall; the mathematical equations of which are as follows: And to calculate the execution time before and after using the cuckoo search, tic and toc Matlab function has been used. The time is calculated from reading the query image until the completion of the comparison between its features vector and those of the database images.
Table (2) illustrates the results of precision, recall and execution time before and after applying cuckoo search.
Table (2): The Results with and without Cuckoo Search Table(2) shows a noticeable difference between the time it takes to retrieve images before and after the Cuckoo search is used. This difference is due to the reduction in the number of features used from (24594)     To evaluate the present proposed system, its obtained results in terms of (precision, recall) have been compared with the results of two previous studies ( table 3 & table 4): The first study by Lakshmi P.S and others (2014) used GLCM to extract texture features from images, and applied Genetic Algorithm to select optimal features, and used Euclidean distance in similarity measure process. The second study by Preeti K. and R. R. Welekar (2016) used color moment, GLCM and edge histogram descriptor to extract features, while the features selection process was done by using Genetic Algorithm and k-means algorithm for clustering. In the current study, the features have been extracted by using color moment ,GLCM and DWT(LL-sub band); features selection has been applied to (LL only) using Cuckoo search algorithm and Manhattan has been used for similarity measure. The comparison has been done on the same image classes from Corel image database. The results (precision & recall) in tables 3 &4 illustrate the efficiency of our proposed system in most image classes. Figure 8 and 9 show the comparison between our proposed system and previous studies in relation to precision and recall.

Conclusion
In the current study, an efficient CBIR system based on extracting multi low-level features like color moment (Mean, Standard Deviation), GLCM, and DWT( only LL-sub band) has been proposed. Cuckoo search algorithm was used to select the important positions that contained power full features from the (LL-sub band). Once a query image was entered, its low-level features were extracted in the same way. Manhattan similarity measure was used to retrieve images from the database related to query mage. It is concluded that the use of the Cuckoo search algorithm to select the features in the proposed system significantly reduced the time required to retrieve images similar to the query image and increased the efficiency of retrieval in terms of precision& recall, compared to the system that was not using the cuckoo search algorithm. Our proposed system was compared with previous works and proved to give good results in terms of precision and recall.