Feature Based Image Sequence Retargeting in the Uncompressed Video Domain

The system propose a video retargeting algorithm to resize images based on the extracted saliency information from the compressed domain. The system utilizes DCT coefficients in JP2 bit stream to perform saliency detection with the consideration of the human visual sensitivity. Valuable retargeting requires emphasize the main satisfied while retain immediate context with minimal visual deformation. A number of algorithms have been proposed for image retargeting with image substance taken as much as potential. But, they usually suffer from deformation results, such as edge or structure twists. A structure and content preserving image retargeting technique is used that preserves the content and image structure. The image content saliency is estimated from the structure of the content using probability map. A block structure energy is use for structure conservation along both directions. Block structure energy uses top down strategy to constrict the image structure consistently. However, the flexibilities of retargeting are altered for different images. To defeat this problem, the patch transform is introduced, where an image is broken into non-overlapping patches, and modifications or constraints are applied in the “patch domain”.. Thus, the resized image is produced to preserve the structure and image content quality.