No abstract uploaded!

We study infinite iterated functions systems (IIFSs) consisting of bi-Lipschitz mappings instead of conformal contractions, focusing on IFSs that do not satisfy the open set condition. By assuming the logarithmic distortion property and some cardinality growth condition, we obtain a formula for the Hausdorff, box, and packing dimensions of the limit set in terms of certain topological pressure. By assuming, in addition, the weak separation condition, we show that these dimensions are equal to the growth dimension of the limit set.

Surface mapping plays an important role in geometric processing. They induce both area and angular distortions. If the angular distortion is bounded, the mapping is called a quasi-conformal map. Many surface maps in our physical world are quasi-conformal. The angular distortion of a quasi-conformal map can be represented by Beltrami differentials. According to quasi-conformal Teichmüller theory, there is an 1-1 correspondence between the set of Beltrami differentials and the set of quasi-conformal surface maps. Therefore, every quasi-conformal surface map can be fully determined by the Beltrami differential and can be reconstructed by solving the so-called Beltrami equation. In this work, we propose an effective method to solve the Beltrami equation on general Riemann surfaces. The solution is a quasi-conformal map associated with the prescribed Beltrami differential. We firstly formulate a discrete analog of quasi-conformal maps on triangular meshes. Then, we propose an algorithm to compute discrete quasi-conformal maps. The main strategy is to define a discrete auxiliary metric of the source surface, such that the original quasi-conformal map becomes conformal under the newly defined discrete metric. The associated map can then be obtained by using the discrete Yamabe flow method. Numerically, the discrete quasi-conformal map converges to the continuous real solution as the mesh size approaches to 0. We tested our algorithm on surfaces scanned from real life with different topologies. Experimental results demonstrate the generality and accuracy of our auxiliary metric method.

A weighted and convex regularized nuclear norm model is introduced to construct a rank constrained linear transform on feature vectors of deep neural networks. The feature vectors of each class are modeled by a subspace, and the linear transform aims to enlarge the pairwise angles of the subspaces. The weight and convex regularization resolve the rank degeneracy of the linear transform. The model is computed by a dierence of convex function algorithm whose descent and convergence properties are analyzed. Numerical experiments are carried out in convolutional neural networks on CAFFE platform for 10 class handwritten digit images (MNIST) and small object color images (CIFAR-10) in the public domain. The transformed feature vectors improve the accuracy of the network in the regime of low dimensional features subsequent to principal component analysis. The feature transform is independent of the network structure, and can be applied without retraining the feature extraction layers of the network.

No abstract uploaded!