In computer vision, what is a filter?

In computer vision, a filter is described as a small grid of pixels that encodes a specific pattern. This small grid is often referred to as a kernel or convolutional filter, and it is used in various image processing tasks. The primary function of such a filter is to convolve with an input image, meaning that it slides over the image, applying the specific pattern it encodes to transform the image data in some way.

For instance, a filter can be designed to highlight edges, smooth an image, or detect certain textures by computing the weighted sum of the pixels in the area covered by the filter. The result of this convolution is a new image that contains different features or enhancements based on the filter's design. This process is fundamental in many computer vision applications, particularly in deep learning models where convolutional neural networks (CNNs) utilize multiple layers of such filters to extract hierarchical features from images.

The other options refer to different concepts in AI and image processing and do not encapsulate the definition and purpose of a filter as accurately as the correct choice does.