A modified Gabor function for content based image retrieval

In-Plane Rotation and Scale Invariant Clustering Using Dictionaries

“…We report the results of empirical evaluation of our method and compare it with six state-of-the-art matching algorithms on three standard datasets: the Smithsonian isolated leaf dataset [32], Kimia's object dataset [31] and Brodatz texture dataset [44]. We compare the performance of our method with a modified Gabor-based approach [21], a local binary pattern (LBP)-based approach [35], and three recently proposed featurebased approaches [30], [32], [33]. We refer to the indexing and retrieval method presented in [33] as the BAC 3 method.…”

“…We evaluate the performance of various methods using precision-recall curves, average retrieval performance [21], [29] and recognition rates. Recall and precision are defined as Precision = Number of relevant images retrieved Total number of images retrieved ,…”

“…For the sub-dataset containing rotated images only, on average RSICD obtained 7.9352 out of 8 retrieved images per shape. Whereas the BAC [33], IDSC+DP [32], SC [30], DC [15], LBP [35] and Gabor-based methods [21] In our CBIR experiments, to determine the class label of a given test image, we find its nth nearest neighbor among the training images, and then assign that training image's estimated class label (given by our RSICD algorithm) to the test image. The nth rank recognition rate is therefore defined as the ratio of the number of test images' nth nearest neighbors (among the training images) that are assigned with the same class labels as the true labels of the test images, to the total number of test images.…”

“…In the experiment section, we introduce both rotation and scale variations to shape-based databases (Smithsonian isolated leaf and Kimia) and compare RSICD with four other state-of-the-art methods (LBP [35], BAC [33], IDSC+DP [32] and SC [30]). We further demonstrate the effectiveness of RSICD on a texture-based database (Brodatz) by comparing it with three other texture-based methods (Modified Gabor [21], DC [15] and LBP [35]). The organization of the paper is as follows.…”

“…However, the chosen directions of Gabor filters may not correspond to the orientation of the content in the query image. Hence, a feature extraction method that is independent of orientation and scale in the image is desirable [21]. Wavelet-based methods have been proposed to achieve rotation invariance for image classification and retrieval [22], [23].…”