Annotated Outdoor Image Compression Based on Sky Replacement

The sky is a basic component in outdoor images. It has high priority in seam removing for content aware image resizing due to the low gradient values of sky regions. The result of enlarging image using the content aware image resizing produces obvious non-natural pixels. This paper presents a technique for Annotated Outdoor Image Compression (AOIC) depending on sky replacing. The outdoor images are compressed by replacing the sky region with the nearest sky in a predefined database of sky images to reduce image size. Our work is composed of two stages: encoding stage and decoding stage. In the encoding stage, The sky annotation polygon is used to distinguish sky and non-sky regions. The sky features are extracted from the sky region (e.g., sky type, mean HSV, cloud type, and sun existence/relative position). These features are used to find the index (SI) of nearest sky feature from a predefined database of sky features. Then, the output image is generated that contains the SI, the original height of the image, the vertices of the sky polygon, and the compressed non-sky image. On the decoding stage, the SI is used to load the corresponding sky from a predefined sky image database and compose it with the non-sky region. Our method is tested using the standard outdoor images of LabelMe database which contain sky. Our method replaces the sky region instead of shrinking the sky region as in content aware image resizing. The AOIC can increase the efficiency of storage for annotated image databases. For example, the outdoor images of LabelMe database which contain sky are 2462 images; the size of these images is 121.5MB. The size of these images after compressing them using the AOIC is 34.1MB. In the security industry, the AOIC can be used to reduce the storage of image database that keeps images containing sky captured from external security cameras. Also, it increases the efficiency of sending these images through a network.