Learning Explicit Smoothing Kernels for Joint Image Filtering

Smoothing noises while preserving strong edges in images is an important problem in image processing. Image smoothing filters can be either explicit (based on local weighted average) or implicit (based on global optimization). Implicit methods are usually time‐consuming and cannot be applied to joint image filtering tasks, i.e., leveraging the structural information of a guidance image to filter a target image. Previous deep learning based image smoothing filters are all implicit and unavailable for joint filtering. In this paper, we propose to learn explicit guidance feature maps as well as offset maps from the guidance image and smoothing parameter that can be utilized to smooth the input itself or to filter images in other target domains. We design a deep convolutional neural network consisting of a fully‐convolution block for guidance and offset maps extraction together with a stacked spatially varying deformable convolution block for joint image filtering. Our models can approximate several representative image smoothing filters with high accuracy comparable to state‐of‐the‐art methods, and serve as general tools for other joint image filtering tasks, such as color interpolation, depth map upsampling, saliency map upsampling, flash/non‐flash image denoising and RGB/NIR image denoising.     

Synthesis of Zirconia Colloids from Aqueous Salt Solutions

We monitor the synthesis of ZrO₂ colloids from dissolved ZrOCl₂·8H₂O using quasi-elastic light scattering and cryo-transmission electron microscopy (cryo-TEM). We investigate the effects of both precursor salt concentration and pH on the final size distribution. We find that the initial pH plays the most important role in the final particle size and yield. These trends and cryo-TEM observations help elucidate the roles of nucleation, aggregation, and recrystallization.     

Modelling the dynamics of log‐domain circuits

Log‐domain filters are an intriguing form of externally linear, internally nonlinear current‐mode circuits, in which a compression stage is first used to convert the input currents to the logarithmic domain, then analogue processing is carried out on the resulting voltages, and finally input–output linearity is restored by mapping the output voltages to current form through an expansion stage. The compressing and expanding operations confer on log‐domain filters a number of desirable features, but they may be responsible for the loss of external linearity. In this paper, sufficient conditions for the external linearity of log‐domain LC‐ladders are established, and the local nature of this external linearity is highlighted. Certain log‐domain LC‐ladders employing floating capacitors may exhibit externally nonlinear behaviour even for zero input and very small initial conditions. We show how transistor parasitic capacitances are central to the emergence of this behaviour, and must be incorporated in the circuit model. Copyright © 2006 John Wiley & Sons, Ltd.     

Learning to classify with missing and corrupted features

A common assumption in supervised machine learning is that the training examples provided to the learning algorithm are statistically identical to the instances encountered later on, during the classification phase. This assumption is unrealistic in many real-world situations where machine learning techniques are used. We focus on the case where features of a binary classification problem, which were available during the training phase, are either deleted or become corrupted during the classification phase. We prepare for the worst by assuming that the subset of deleted and corrupted features is controlled by an adversary, and may vary from instance to instance. We design and analyze two novel learning algorithms that anticipate the actions of the adversary and account for them when training a classifier. Our first technique formulates the learning problem as a linear program. We discuss how the particular structure of this program can be exploited for computational efficiency and we prove statistical bounds on the risk of the resulting classifier. Our second technique addresses the robust learning problem by combining a modified version of the Perceptron algorithm with an online-to-batch conversion technique, and also comes with statistical generalization guarantees. We demonstrate the effectiveness of our approach with a set of experiments.     

Developing a Combustion Model of a Solid Propellant Containing Capsules of Liquid Oxidizer

A combustion model of a solid propellant containing liquid oxidizer capsules was developed. The model is based on the combustion cycle of a unit cell, an oxidizer droplet surrounded by the adjusting amount of binder. Three combustion stages are considered: (1) binder decomposition parallel to oxidizer heating up to boiling temperature, (2) simultaneous binder and oxidizer gasification, and (3) gasification of the remaining condensed phase species. Simple global kinetics are assumed for the gas phase reactions and binder decomposition abiding Arrhenius law, and boiling process for the oxidizer abiding Clausius‐Clapeyron curves. The obtained results show similar trends as do other relevant models and experimental results, especially for the effect of droplet size and propellant composition on the burn rate. The predicted effect of initial temperature on the burn rate is less significant than for common solid propellants. The results indicate that for small droplets the oxidizer will heat up to boiling temperature even before it is revealed on the burning surface, due to heat conduction through the surrounding binder.     

Investigation of Slow-Propagation Tungsten Delay Mixtures

The burning characteristics and reliability of slow-propagation tungsten delay mixtures (characteristic time 38–42 s/in) were studied experimentally by using various mixture compositions and tungsten particle sizes. Thermal analysis showed that potassium perchlorate is vital to initiate the reaction. Gravimetric analysis indicated that during the reaction overall mass loss of the delay mixtures did not exceed 5% compared to their initial mass. A mixture containing a tungsten powder of 6–8 μm was found to have the lowest effective activation energy, facilitating good ignition and burning zone propagation. The results of this study were successfully used to design a tiny delay which exhibited reliable ignition, burning and propagation characteristics under extreme conditions.     

Universal lossless source controlled channel decoding for i.i.d. sequences

Messages coded and transmitted over a channel usually contain some redundancy which is not utilized by channel decoding techniques, especially if its governing statistical parameters are unknown. We propose to integrate universal lossless source coding techniques into channel decoding of redundant sequences with unknown statistics to improve performance of Viterbi and turbo decoding. Simulation results demonstrate that we achieve identical bit error rate (BER) performance to nonuniversal techniques that utilize prior knowledge of the message statistics and that if redundancy exists even in an a-priori unknown form, we can improve the code performance over standard techniques.