Image restoration refers to the recovery of an underlying image from an observation corrupted by various types of noise. In digital forensic software, such an image restoration process should be noise-tolerant, robust, fast, and scalable. Among many e...
Image restoration refers to the recovery of an underlying image from an observation corrupted by various types of noise. In digital forensic software, such an image restoration process should be noise-tolerant, robust, fast, and scalable. Among many existing models of associative memory, Fuzzy Associative Memory (FAM) is widely used in the implementation of such systems using a fuzzy Hebbian learning rule. Such methods are often based on max-min or max-product compositions for the synthesis of the weight matrix. The FAMs provide important advantages including noise tolerance, unlimited storage, and one-pass convergence nevertheless they have low capacity. FAM performance is also related to its ability to capture the content of each pattern and its association. Therefore, when the traditional FAM is applied to recover images, the recovery rate of images is high for the same individuals with different backgrounds, but low if the same individuals have the same background. To address this issue we propose a T_norm-based FAM technique. In fuzzy theory, the operations that can be applied to the value of the function to which they belong are T_norm (Triangular-norm) and T_conorm (Triangular-conorm). Here we apply the T_norm to improve the degree of repair for the FAM weighted value operation. To verify the performance of the proposed T_norm-based FAM method, we conduct experiments with 20 images. In our dataset, researchers conduct experiments with 20 images of objects damaged against the same background and 20 images of objects damaged against different backgrounds were tested.. For both cases, 90% recovery performance is achieved using the T_norm-based FAM method. This shows % recovery performance in the existing FAM. Nevertheless, the RMSE is high with the proposed T_norm-based FAM method and the existing FAM all recovered in the video.