Image Hashing Algorithms Resistant To Rotation

Image hashing is a novel research topic of digital media content processing, which has been widely applied to many fields, such as image authentication, digital watermarking, image retrieval and image copy detection. It uses a short string of characters/digits to denote an image. In practice, digital images often undergo.some digital processing. After these processing, visual appearances between original and processed images are unchanged, but digital representations are quite different. Consequently, image hashes of the original and processed images are expected to be unchanged or very similar. This is the first property of image hashing called perceptual robustness. Another basic property of image hashing is called discrimination. It means that digital images with different contents should have different hashes. Rotation is one of the basic image operations, which plays an important role in many applications, such as image recognition, image correction, image stitching, and processing the remote sensing images. However, most of the existing hashing algorithms can not reach an acceptable trade-off between rotation robustness and discrimination. This means that the rotated image will be falsely classified as different images. To solve this problem, this thesis investigates image hashing algorithms resistant to rotation. Specifically, I first analyzed the theory of image rotation, and then proposed a ring-based image division method and three image hashing methods. Specific research results are illustrated as follows.1.1proposed a ring-based image division method.I first analyzed the theory of image rotation, and then proposed a ring-based image division method. This method divides an image into different rings, and the contents of these rings are rotation-invariant. I found and proved the property that image pixels of each ring are almost unchanged after image rotation. This property is a fundamental theory for designing ring-based image hashing.2.1proposed a robust image hashing based on multiple histograms.This image hashing is done by converting the input image into a normalized image, dividing it into different rings, extracting ring-based histograms and compressing them by discrete wavelet transform. As ring-based histograms are kept unchanged after rotation, the extracted histograms are rotation-invariant and then the proposed algorithm is robust against rotation with arbitrary angle and has good discriminative capability.3.1proposed an image hashing based on color vector angles and Canny operator.Edge is an important visual feature for human visual system to distinguish different images, and will not be significantly changed by normal digital processing. Based on these observations, I proposed an image hashing with image edges, color vector agnles and ring-based division. This algorithm firstly detects image edges by Canny operator, calculates color vector angles of input image, selects a set of circles, and calculates variances of color vector angles of edge points on the circles to construct image hash. This method achieves rotation robustness by using rotation-variant property of circle.4.1proposed a robust image hashing based on fan-beam transform.Fan-beam transform is a variation of Radon transform. It inherits excellent properties of Radon transform, but runs faster than Radon transform. This hashing converts an input image into a normalized image by bi-cubic interpolation and color space conversion, applies fan-beam transform to the normalized image, and exploits those variances of discrete Fourier transform coefficients in fan-beam projections. to construct image hash. Rotation robustness of this algorithm is provided by the rotation-invariant property of fan-beam transform.Many test images are exploited to validate perceptual robustness and discriminative capability of the above three algorithms. Experiments illustrate that the proposed algorithms are all robust against normal digital processing, such as JPEG compression, brightness adjustment, contrast adjustment, gamma correction, image scaling and image rotation. Receiver operating characteristics (ROC) curve comparisons with some existing popular hashing algorithms are also conducted, and the results demonstrate that the proposed algorithms have better performances than the compared algorithms in classification between robustness and discrimination.