Image Hashing Based On Compressive Sensing

Image hashing is a new technology in the field of multimedia information security, which has been widely used in image authentication, digital watermarking, multimedia indexing, image copy detection, image retrieval and so on. There are two basic properties for image hashing, namely, robustness and uniqueness. The similar images, regardless of their specific data representation, should be generated the same or similar hashes. This is a significant difference between image hashing algorithm and traditional cryptography hashing algorithm. The traditional cryptographic hashing algorithm can map any input data into a fixed-size string, even only one bit change of the input data will lead to dramatic changes. Image were commonly applied a series of operations in the daily application, such as image enhancement, image compression, geometric transformation and so on. These operations will change the image of the specific representation of data, but does not change the visual content of the image, so the image hash values supposed unchanged. Hence, cryptography hash algorithm is only suitable for text data and for multimedia data, we need to design and develop a new hash algorithm for digital image. On the other hand, the hash values of images have different contents, should be entirely different, i.e., image hash has good uniqueness. In addition, image hash, which applied in the areas of image authentication, should be safe. In other words, for the same image, different keys generated hash value should be completely different.This thesis chooses the digital image for the research, investigates compressed sensing and color vector angle, presents two kinds of image hashing algorithm, i.e., image hashing based on color vector angle and compressive sensing and image hashing based on compressive sensing and ring partition.The detailed research content is as follows:1. I proposed an image hashing algorithm based on compressive sensing and color vector angle.Compressive sensing is an effective dimensionality reduction method, which can map data into a low dimensional space by measurement matrix and extract the image feature to construct hash function. The color vector angle of brightness change is not sensitive for the color image, but the hue and saturation is sensitive. Compared with the Euclidean distance, evaluation of color vector angle between two difference colors image is more effective, can capture the color perception differences. In general, the input image undergoes pre-process to establish a standardized image, laying the foundation for extracting robust features. Then the color vector normalized image and the corresponding angle matrix has been calculated. Finally, the matrix block separately has been compressed sensing. Experimental results show that the proposed algorithm can resist digital normal operation, such as JPEG compression, watermark embedding, brightness/contrast adjustment, gamma correction, Gaussian low pass filtering, scaling and small angle rotation.2.1 proposed an image hashing algorithm based on compressed sensing and ring partition.As most image hashing algorithm is not robust for large angle rotation operation and the ring partition compared to the rectangular block can resist rotation operation, I designed an image hashing algorithm to resist rotation operation. First of all, the input image was pre-processed. Secondly, image was partition into several rings and pixels of each ring are extracted. Next, these ring pixels are compressed with compressive sensing. Finally, the image hash is obtained by calculating inner product between measure vector and random vector. Experimental results show that the proposed algorithm can resist digital normal operation, such as JPEG compression, scaling, brightness/contrast adjustment, and large angle rotation. Receiver operating characteristic (ROC) curve comparisons illustrate that the proposed algorithm is better than FM-CS hashing, GF-LVQ hashing and RT-DCT hashing in classification performance.