Abstract

In general, a neural network is a device to simulate the brain. Neural network theory revolves around the idea to derive and add to simulations the core properties of biological neurons, thereby constructing a simulated (and simplified) brain. An ANN is developing through a learning process for a given application, such as pattern recognition or data classification. Training in biological systems requires modification of the synaptic connections between the neurons. For block-based image / video compression systems, blocking artifacts, characterized by visually noticeable changes for pixel values along block boundaries, may be a fundamental issue. Different post-processing methods are introduced to scale back blocking artifacts, but most usually add unnecessary blurring or ringing effects. This study provides a process for estimating cardiac motion in 2-D ultrasonic images. The problem of motion estimation is formulating as an energy minimization, of which the term of data fidelity is built on the belief that the photographs are corrupted by multiplicative Rayleigh noise. In addition to some kind of classical spatial smoothness constraint, the proposed solution takes advantage of the sparse properties of cardiac motion to regularize the response through a suitable dictionary learning stage. The proposed approach is tested on one set of data with available ground-truth like four highly realistic simulations sequences.