Markus Juvonen

The initial spark for this work was a thesis project, exploring the potential of patch-based image processing. The visually interesting results inspired me to pursue a more artistic development of the technique. The core is an algorithm programmed using MATLAB, that breaks a set of images into patches, and uses k-means clustering to group similar patches together. Then patches of a reference image are replaced by members of the best matching group, according to a selected distance metric, creating a new visual narrative. Incorporating an element of randomness in the selection process for each representative of patches, in each frame of the video, creates a sense of movement and fluidity that is visually captivating. The residual, obtained by subtracting the processed image from the original, visualizes algorithm performance and the relation of the images. By reconstructing images from entirely different images I invite viewers to consider new ways of visualizing the world around us.