Recognises the human silhouette in the input image, removes the background and returns the foreground portrait image (4 channels). Suitable for scenes with single/multiple people, complex backgrounds, various types of human poses, etc.