Content-Aware-Image-Resizing

Seam Carving Algorithm for content aware image resizing and object removal

Importance/Cost Matrix

two types of cost functions are implemented

• computes image gradient as cost
• computes cost as per pixel entropy considering 9-by-9 neighbourhood

original image	gradient cost

Image Downsizing

• does reduction in width using seam carving
• results below are for gradient cost matrix

original width : 1428 pixels	reduced width : 1128 pixels	reduced width : 714 pixels	reduced width : 476 pixels

• per pixel entropy generally works better than gradient but it is computationally expensive

original width : 470 pixels	gradient cost, reduced width : 235 pixels	entropy cost, reduced width : 235 pixels

Image Upsizing

• does increment in width using seam carving
• if width is to be increased by x pixels, then width is reduced by x pixels first using normal seam carving and the deleted seams are duplicated in original image to get the resultant image with width increased by x pixels

original width : 1428 pixels	increased width : 2142 pixels

original width : 464 pixels	increased width : 696 pixels

original width : 520 pixels	increased width : 2128 pixels

Object Removal

• resize the image in such a way that specific object in the image is removed smoothly
• bounding box over the object is given and the cost values of the pixels in the bounding box are made negative infinite to make every seam pass through the object

original	object removed

original	object removed