In today’s applications of consumer and industrial electronics image and video processing have become more and more important in achieving, respectively, either a more interactive user experience or a more efficient use of electronic control systems: the need for real time video processing arises from these possibilities in order to, for example, allow the implementation of computer vision algorithms. Applications may range from industrial processes control to object recognition, automatic navigation systems or image-driven information retrieval. In this paper we present a dedicated coprocessor for frequency domain filtering of block-partitioned frames (which may come either from a single image or a video stream) realized with a scale-space oriented approach, accordingly to [1]. The particular filtering implemented in the presented architecture realizes the Difference of Gaussians algorithm: four bandpass filters are realized by means of subtracting pairs of adiacent Gaussian lowpass filters. The block partitioning of frames performed prior to the Fast Fourier Transform, as proven in [2], can reduce the amount of operations required to perform the filtering operations while providing flexibility regarding the image/frame size.
An FFT-based coprocessor for real time video processing
Ettore Napoli
2012-01-01
Abstract
In today’s applications of consumer and industrial electronics image and video processing have become more and more important in achieving, respectively, either a more interactive user experience or a more efficient use of electronic control systems: the need for real time video processing arises from these possibilities in order to, for example, allow the implementation of computer vision algorithms. Applications may range from industrial processes control to object recognition, automatic navigation systems or image-driven information retrieval. In this paper we present a dedicated coprocessor for frequency domain filtering of block-partitioned frames (which may come either from a single image or a video stream) realized with a scale-space oriented approach, accordingly to [1]. The particular filtering implemented in the presented architecture realizes the Difference of Gaussians algorithm: four bandpass filters are realized by means of subtracting pairs of adiacent Gaussian lowpass filters. The block partitioning of frames performed prior to the Fast Fourier Transform, as proven in [2], can reduce the amount of operations required to perform the filtering operations while providing flexibility regarding the image/frame size.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
