Visual Security Index Combining CNN and Filter for Perceptually Encrypted Light Field Images

Visual security index (VSI) represents a quantitative for the visual evaluation of perceptually encrypted images. Recently, research on light field (LF) images faces two challenges. One is that existing image databases are often too small, which easy to cause overfitting in convolutional neural network (CNN). The other VSI models did not take full account intrinsic characteristics LF and highly relied handcrafted feature extraction. In this paper, we construct new database images, called PE-SLF, 2.6 times as big largest perceptual database. Moreover, novel model proposed by taking into consideration spatial-angular outstanding capabilities CNN First, exploit detect texture structure features sub-aperture spatial domain. Second, apply Gabor filter over epi-polar plane angular Last, similarity measurements subsequently calculated jointly yielding final score. Experimental results constructed PE-SLF demonstrate closer perception HVS compared classical state-of-the-art models.