ABSTRACT Deep learning models require large training datasets. Incorporating additional data into small training datasets can enhance the model's performance. However, acquiring additional data may sometimes be challenging or beyond one's control. In such situations, data augmentation becomes essential to overcome the limited supply of labeled data by generating new data that preserves the essential properties of the original dataset. The primary objective of our research is to develop an iterative data distillation and augmentation (IDDA) method that enlarges the size of a limited image training dataset while preserving its properties. At every iteration, our method distills a set of images from the training set of the previous iteration utilizing the kernel inducing point (KIP) method, and the union of the training and distilled sets creates the new training set. However, our experiments show that IDDA is computationally expensive, increasing processing time by approximately 17%–27 for MNIST and Fashion‐MNIST, 31%–39 for CIFAR‐10, and up to 48%–49 for CIFAR‐100 compared to state‐of‐the‐art augmentation methods, due to the additional step of applying KIP for image distillation. We have experimentally determined that for a few iterations the classification accuracy increases and then drops afterward. We validate the IDDA capabilities by comparing it with conventional augmenting methods and MixUp on the following publicly available image datasets: MNIST digit, Fashion‐MNIST, CIFAR‐10, and CIFAR‐100. Our approach proves highly effective for very limited datasets, addressing the challenge of database expansion for improved performance of deep learning models.

Active contours are very popular tools for video tracking and image segmentation. Parameterized contours are used due to their fast evolution and have become the method of choice in the Sobolev context. Unfortunately, these contours are not easily adaptable to topological changes, and they may sometimes develop undesirable loops, resulting in erroneous results. To solve such topological problems, one needs an algorithm for contour self-crossing detection. We propose a simple methodology via simple techniques from differential topology. The detection is accomplished by inspecting the total net change of a given contour's angle, without point sorting and plane sweeping. We discuss the efficient implementation of the algorithm. We also provide algorithms for locating crossings by angle considerations and by plotting the four-connected lines between the discrete contour points. The proposed algorithms can be added to any parametric active-contour model. We show examples of successful tracking in real-world video sequences by Sobolev active contours and the proposed algorithms and provide ideas for further research.

Visual tracking of arbitrary targets in clutter is important for a wide range of military and civilian applications. We propose a general framework for the tracking of scaled and partially occluded targets, which do not necessarily have prominent features. The algorithm proposed in the present paper utilizes a modified normalized cross-correlation as the likelihood for a particle filter. The algorithm divides the template, selected by the user in the first video frame, into numerous patches. The matching process of these patches by particle filtering allows one to handle the target's occlusions and scaling. Experimental results with fixed rectangular templates show that the method is reliable for videos with nonstationary, noisy, and cluttered background, and provides accurate trajectories in cases of target translation, scaling, and occlusion.

We propose a tracking system that is especially well-suited to tracking targets which change drastically in size or appearance. To accomplish this, we employ a fast, two phase template matching algorithm along with a periodic template update method. The template matching step ensures accurate localization while the template update scheme allows the target model to change over time along with the appearance of the target. Furthermore, the algorithm can deliver real-time results even when targets are very large. We demonstrate the proposed method with good results on several sequences showing targets which exhibit large changes in size, shape, and appearance.