Adaptive Noise Filtering of Image Sequences in Real Time

Filtering noise in image sequences is an important preprocessing task in many image processing applications, including but not limited to real-time x-ray image sequences obtained in angiography. The main objective in real-time noise filtering is to improve the quality of the resultant image sequences. Practically affordable approaches are generally suboptimal and deal with the spatial and temporal dimensions independently. Spatial filters can be adaptive and edge sensitive, however, they may require more hardware real estate for the real-time processing of each frame. On the other hand, temporal-only filters are one-dimensional and take advantage of temporal correlation. These 1-D temporal filters, which are applied to each individual pixel, can be designed using adaptive approaches to compensate for motions as well as noise variations. Existing adaptive 1-D filters are relatively complex and do not lend themselves to an affordable hardware implementation for real-time processing. In this article, after reviewing different filtering approaches, an adaptive temporal restoration algorithm, based on discrete Kalman filter, is developed. Adaptation in this case is with respect to the variation of the noise statistics as well as motion. In each step of the algorithm, the conventional adaptive Kalman filter proceeds if no motion is detected. However, in the case of detected motion, the adaptive Kalman filter resets itself in a way that the motion is preserved and cause no lagging in the processed image sequence. The overall procedure is suitable for hardware implementation with present FPGA/VLSI technology. Key–Words: Image Sequence Filtering, Temporal and Spatial Filtering, Adaptive Filtering, Kalman Filter