Artificial Intelligent Systems and Machine Learning

Hanan H. Elazhary ¹, Sawsan M. Gharghory ¹

Affiliations
1 Computers and Systems Department, Electronics Research Institute, Cairo, EG

Abstract

Particle Swarm Optimization (PSO) algorithm is used in the literature to solve different types of optimization problems. Most PSO algorithms suffer from the problem of the long processing time and the fact that PSO gets trapped easily in a local minimum. This paper argues the need to address both problems simultaneously to enhance the performance of PSO unlike other research studies in the literature that typically address only one of these problems. To emphasize this, the paper proposes to combine the basic PSO algorithm with both evolutionary operators and parallel processing to solve the rounding error problem of a DCT-based fragile image watermarking algorithm. Evolutionary operators help PSO jump out of local minima, while parallel processing helps speed up processing. In such watermarking algorithms, the watermark bits are typically embedded by modifying the least significant bits of the frequency coefficients of the DCT-transformed host image. The host image is then converted into the spatial domain using inverse DCT and the real numbers of the inverse-DCT coefficients are rounded to integers. This enables detecting the slightest modification of the host image, but the rounding process results in a significant difference between the extracted watermark and the embedded one and reduces the watermarked host image quality. The proposed algorithm is compared to three cases: a) using PSO alone, b) using PSO with evolutionary operators without parallel processing, and c) using PSO with parallel processing and without evolutionary operators. The experimental results show the superiority of our approach.

Keywords

Cauchy Mutation, Evolutionary Operators, Fragile Image Watermarking, Parallel Processing, Particle Swarm Optimization.

Full Text

     

Hanan H. Elazhary ¹, Heba A. El Nemr ¹

Affiliations
1 Computers and Systems Department, Electronics Research Institute, Cairo, EG

Abstract

Video abstraction or summarization is a technique that generates a short summary of the video possibly as a group of stationary keyframes. Video abstraction has several applications including video content browsing and visualization. The goal of this research study is to develop a comprehensive approach for generating different sets of keyframes according to the application while speeding up the processing time and minimizing human intervention. The proposed approach relies on dividing the video into equal short segments to avoid the delay resulting from shot segmentation. Keyframes are then generated from these segments using Particle Swarm Optimization (PSO) for additional speedup. Since different runs produce different sets of keyframes, automatic integration of the results of few runs has been shown to produce a set of keyframes capable of capturing the changes within each video shot for video content browsing without the need for human intervention. Since this implies the increase of the overhead, parallel processing is utilized for further significant speed up. The generated keyframes are filtered to produce a satisfactory set of keyframes for video visualization while minimizing the false rate, increasing the hit rate, and minimizing human intervention.

Keywords

Parallel Processing, Particle Swarm Optimization, Video Content Browsing, Video Summarization, Video Visualization.

Full Text