An enhancement of mini-saic algorithm applied in photomosaic

ABSTRACT: Photomosaic is a procedure that transforms as an image into a rectangular grid of thumbnail while resembling its original appearance. The typical photomosaic algorithm searches from a large database of images for one picture that resembles the closest a tile in a target image. Three problems were seen and experienced in the MiNi-Saic’s Photo Mosaic Algorithm and the application itself; (1) Pinpointing cannot assume the exact boundary of the subject of the photo, (2.a) Target image varies in size therefore tile size may not fit the target image, (2.b) Top-left most portion alone of dissimilar dimensioned source image is acquired in the computation of its RGB values, (3) A tile may contain details the RGB mean may forsake on tile matching, (4) Source image is barely recognizable even when photomosaic output is enlarged, (5) Some source images may not appear on the mosaicked image. Enumerated problems were solved by the proponents by (1) Providing a precise block selection of main subject on target image, (2.a) Insuring as to whether tile size is just fitted with the dimensions of the target image, (2.b) Insuring that accurate computation on the RGB values of source images of any size is performed, (3) Enhancing the tile-matching procedure by getting RGB mean of each image and each tile per region, (4) Providing a better scale of source images so that tiles of the output image, when enlarge, should the small images patched onto it be recognized, (5) Providing an option for the user if he prefers that each of his chosen source images must have at least one instance on the resulting mosaic.

ABSTRACT: Photomosaic is a procedure that transforms as an image into a rectangular grid of thumbnail while resembling its original appearance. The typical photomosaic algorithm searches from a large database of images for one picture that resembles the closest a tile in a target image. Three problems were seen and experienced in the MiNi-Saic's Photo Mosaic Algorithm and the application itself; (1) Pinpointing cannot assume the exact boundary of the subject of the photo, (2.a) Target image varies in size therefore tile size may not fit the target image, (2.b) Top-left most portion alone of dissimilar dimensioned source image is acquired in the computation of its RGB values, (3) A tile may contain details the RGB mean may forsake on tile matching, (4) Source image is barely recognizable even when photomosaic output is enlarged, (5) Some source images may not appear on the mosaicked image. Enumerated problems were solved by the proponents by (1) Providing a precise block selection of main subject on target image, (2.a) Insuring as to whether tile size is just fitted with the dimensions of the target image, (2.b) Insuring that accurate computation on the RGB values of source images of any size is performed, (3) Enhancing the tile-matching procedure by getting RGB mean of each image and each tile per region, (4) Providing a better scale of source images so that tiles of the output image, when enlarge, should the small images patched onto it be recognized, (5) Providing an option for the user if he prefers that each of his chosen source images must have at least one instance on the resulting mosaic.

Filipiniana

5