A Close inspection of the progress made in the field of image processing in the past several decades reveals that
much of it is a direct consequence of the better image modeling employed. Armed with a stronger and more reliable model, one can better handle applications ranging from sampling, denoising, restoration, and reconstruction in inverse problems, all the way to compression, detection, separation, and beyond. Indeed, the evolution of models for visual data is at the heart of the image processing literature. What is a model and why do we need one? We provide an initial answer to these questions through a simple example of noise removal from an image. Given a noisy image, a denoising algorithm is essentially required to separate the noise form the (unknown) clean image. Such a separation clearly requires a close familiarity with the characteristics ofboth the noise and the original image. Knowing that the noise is additive, white, and Gaussian (AWG) is a good start, but far from being sufficient, since the underlying image may also behave like such noise, thereby making the separation of the two impossible. The additional information on the clean image content, that will allow separating it from the AWG noise, constitutes what we refer to in this paper as an image model. A classic example of such a model is the intuitive assumption
A Close inspection of the progress made in the field of image processing in the past several decades reveals that
much of it is a direct consequence of the better image modeling employed. Armed with a stronger and more reliable model, one can better handle applications ranging from sampling, denoising, restoration, and reconstruction in inverse problems, all the way to compression, detection, separation, and beyond. Indeed, the evolution of models for visual data is at the heart of the image processing literature. What is a model and why do we need one? We provide an initial answer to these questions through a simple example of noise removal from an image. Given a noisy image, a denoising algorithm is essentially required to separate the noise form the (unknown) clean image. Such a separation clearly requires a close familiarity with the characteristics ofboth the noise and the original image. Knowing that the noise is additive, white, and Gaussian (AWG) is a good start, but far from being sufficient, since the underlying image may also behave like such noise, thereby making the separation of the two impossible. The additional information on the clean image content, that will allow separating it from the AWG noise, constitutes what we refer to in this paper as an image model. A classic example of such a model is the intuitive assumption
การแปล กรุณารอสักครู่..
