An Effective Hybrid Framework Based on Combination of Color and Texture Features for Content-Based Image Retrieval

In Content-Based Image Retrieval (CBIR) system image representation is a vital function. In the CBIR method, shape, texture, and color features are employed to represent an image for retrieval purposes. Color and texture features are the most remarkable features for defining image content. The combination of these features does not however lead to better retrieval accuracy because of transformation factors such as scaling, rotation, and translation of an image. Another concern regards memory space that affects the running time during image retrieval. This paper thus addresses these concerns by proposing an effective hybrid framework based on the combination of texture and color features by extracting vast important and robust features from a database of images and storing these features as feature vectors. The proposed CBIR framework is based on Color Moments (CMs), Color Histogram, Ranklet Transform, RGB and HSV color spaces; and hybrid Graph-based Gray-Level Co-Occurrence Matrix (HGGLCM), where CMs and Ranklet transforming methods are employed to extract the color features and HGGLCM method was employed to extract the texture features. These descriptors combined enhance image retrieval framework performance. Support Vector Machine Classifier was applied to classify the image database and Squared Euclidean Distance Measurement for displaying the retrieved images. Corel-1k and Oxford Flower datasets were used to test the CBIR methods based on color and texture analysis, the results showed that this proposed framework performed better based on recall and precision values compared with current state-of-the-art CBIR methods, where the proposed method obtained an average precision of 92 and an average recall of 18.4 based on Corel-1k dataset, we obtained an average precision of 0.917 and an average recall of 0.223 based on Oxford Flower dataset.