Image Cropping Based on Saliency and Semantics

This thesis proposes a new automatic image cropping technique and a platform for subjective image quality evaluation on mobile devices. Image cropping is a widely used technique in the printing industry, photography and cinematography. The proposed cropping method considers both the low-level pixel properties and high-level semantics. It is a combination of saliency-based and semantics-based image analysis. In the end, we compare the proposed method with a conventional saliency-based strategy. Furthermore, in order to simplify the final subjective test, we developed an iOS based mobile application for subjective image quality evaluation. The developed application implements two-alternative forced choice (2AFC) test methodology and further reduces the cognitive load of subjects performing the test by providing an easy-to-use, natural interface using the mobile device’s touch screen. The test results show the proposed cropping technique performs significantly better overall compared to saliency-based cropping.