Often times, raw data is redundant and have patterns. Thus, a more efficient representation of the data would be useful. One way to do this is singular value decomposition (SVD). SVD is an algorithm that factors any matrix A into three matrices UΣVT, where U and V are rotation matrices and Σ is a scaling matrix. One of its applications is image compression. SVD of an image matrix produces components of the final image.
As a result, the minimal representation of images and data samples by including only its major patterns with this technique, known as Principal Component Analysis, increases the speed of machine learning algorithms significantly. To learn more about PCA and how it can be used for face detection, click here.
We have all heard of image filters in applications like Instagram. Well, computer vision is used to create the enhancing effects! Different filters produce different effects. For example, the moving average filter smoothes an image, while other manipulations can sharpen or denoise an image. To learn more about systems and filters, click here.
Image matching has several applications, such as panorama stitching. To effectively align the photos, keypoints are identified, a region around each keypoint is defined and normalized, and a local descriptor is computed from the extracted region. The local descriptors are then matched to stitch images together. Keypoint detection can also be used to recognize specific objects and scenes. To learn more about key points and different detection algorithms, click here.
Sometimes, it can be hard to analyze the raw pixels of an image. This is when segmentation comes in handy. Image segmentation is the clustering of pixels into regions based on surfaces, objects, or parts of objects. The divided regions are then represented as a single entity, simplifying the image without losing content. Parts of an image can also be taken and put onto another image. For more information on segmentation, clustering, and some common algorithms used to perform this, click here.
We often overlook the huge role object recognition plays in our daily lives. Animals rely on it to survive, and it is essential for us as well. To learn more about object recognition, click here.
A video consists of many images captured over time. Optical flow is the pattern of motion of the objects, surfaces, or contours in a scene. The estimation of optical flow between frames can give the velocities of the various elements in the video and be used to track visual features over the video frames. To learn more about feature tracking, click here.