In this study, an affine registration algorithm with reference-omitted scheme and soft correspondence is proposed. It is an iterative method with two-step matching process at each iteration, named as forward matching and backward matching. Due to the introduction of backward matching, two sets of points are alternately to be reference set, such that the selection of reference set is omitted. Failure caused by different reference sets can be corrected with the reference-omitted scheme, and even there is obvious difference in scale. Additionally, soft correspondence is applied to avoid estimating the initial transformations. The simulation and real experimental results show that the proposed method substantially outperforms the current affine registration methods, especially when the scale difference between two sets of points is obvious or there are outliers in one set.

A new computationally efficient algorithm for two-dimensional sliding-window least-squares prediction is presented in this study. The fast algorithm is based on a recursive update of the Cholesky decomposition. Compared with the state-of-the-art algorithm, the proposed algorithm reduces the computational complexity from O(D ³) to O(D ² h), where D is the predictor order and h is the height of the prediction patch. The computational improvement is made at the stage of solving the normal equations for which an update algorithm for the Cholesky decomposition of the covariance matrix is proposed. It is shown that a large part of the Cholesky decomposition at location n can be efficiently calculated by performing orthonormal updates on the Cholesky decomposition at n − 1. The computational improvement is made without requiring additional storage space. Extensive experiments using causal and non-causal predictors of varying shapes and sizes have confirmed that the proposed algorithm is consistently faster than the state-of-the-art algorithm and produces identical prediction images. The efficiency of the proposed algorithm is shown to be affected by the order in which pixels are sampled, thus an ordering procedure is proposed to minimise the number of numerical operations.

In this study, the authors propose a novel (2, 2) secret image sharing scheme in which a control parameter ω is used to change the payload easily. Since the modification of the original cover pixel values can be limited within a small range according to the value of ω, the shadow images can achieve excellent visual quality. In the extracting process, the secret image and the cover image can be reconstructed correctly. Experimental results showed that the authors’ proposed scheme can enhance the embedding rate significantly, up to 3 bpp if ω is set to 6. In addition, the peak signal-to-noise ratio values of the shadow images are still satisfactory when the embedding rate approaches a very high value. Comparisons demonstrated that their proposed scheme outperforms other schemes that have been developed recently in terms of the embedding rate and the visual quality.

Sparse representations of multi-channel signals have drawn considerable interest in recent years. In this study, a new vector-valued sparse representation model is proposed for colour images using reduced quaternion matrix (RQM). The colour image is described as a RQM by the proposed model. In the dictionary training state, k-means clustering RQM value decomposition is proposed which makes sparse basis selection in quaternion space. Then, a reduced quaternion-based orthogonal matching pursuit algorithm is presented in the sparse coding stage. To demonstrate the effectiveness of the proposed sparse representation model, the authors apply the model to common colour image processing problem-colour image denoising. The proposed model is compared with other sparse models for colour image denoising in terms of visual quality and peak signal-to-noise ratio. The experimental results indicate that the proposed image sparse model is competitive with other sparse models.

In this study, the authors present a clustering-based thresholding technique for image segmentation. This technique is built on the minimum within-class variance of a scalable local region that draws upon the previous result and its spatial information to account for the connectivity between the background and the object. The cluster mean derived from the object region in each iteration is considered as an alternative global threshold to prevent the pixels with low intensity from clustering and enable the pixels with similarity to be clustered. This approach makes the method less sensitive to the problem associated with the shape of the histogram and thus leads to an automatic iterated method for a promising segmentation performance. Experiment results on synthetic and real images prove the efficiency of the method.

This study models facial expression recognition with a sparse representation classification (SRC) method. By analysing SRC's robustness to noise, this study further proposes an SRC method based on positive and reverse templates (PRTs-SRC), which uses PRTs to expand an over-complete dictionary constructed by training samples. The expanded dictionary can contain more information, and increase the robustness to noise. To validate the performance of the proposed algorithm, experiments were carried out on relevant expression databases. The authors compared and analysed the recognition performances for the proposed algorithm and other methods. The results show that even with high noise levels, the proposed algorithm performs above 80% recognition rate.

Various modes of transmission electron microscopy can provide unique structural information on nanometre-scale biological machines such as viruses and ribosomes. A maximum-likelihood statistical reconstruction algorithm using information from two such modalities, single-particle cryo-electron microscopy and computed electron tomography, is described for the problem where the machine is nearly symmetrical, but the localised regions of asymmetry are important for the functioning of the machine. The algorithm is demonstrated on experimental bacteriophage Lambda procapsid data. Key features of the algorithm are the use of probability density functions (pdfs) derived from normalised correlation rather than Gaussian pdfs and the solution of a constrained classification problem in which 11 out of 12 sites of asymmetry belong to one class while the 12th site belongs to a second class.