The process of image retrieval presents an interesting tool for different domains related to computer vision such as multimedia retrieval, pattern recognition, medical imaging, video surveillance and movements analysis. Visual characteristics of images such as color, texture and shape are used to identify the content of images. However, the retrieving process becomes very challenging due to the hard management of large databases in terms of storage, computation complexity, temporal performance and similarity representation. In this paper, we propose a cloud-based platform in which we integrate several features extraction algorithms used for content-based image retrieval (CBIR) systems. Moreover, we propose an efficient combination of SIFT and SURF descriptors that allowed to extract and match image features and hence improve the process of image retrieval. The proposed algorithms have been implemented on the CPU and also adapted to fully exploit the power of GPUs. Our platform is presented with a responsive web solution that offers for users the possibility to exploit, test and evaluate image retrieval methods. The platform offers to users a simple-to-use access for different algorithms such as SIFT, SURF descriptors without the need to setup the environment or install anything while spending minimal efforts on preprocessing and configuring. On the other hand, our cloud-based CPU and GPU implementations are scalable, which means that they can be used even with large database of multimedia documents. The obtained results showed: 1. Precision improvement in terms of recall and precision; 2. Performance improvement in terms of computation time as a result of exploiting GPUs in parallel; 3. Reduction of energy consumption.

This paper introduces an automated system designed to recognize specifically speech in the Amazigh language, using a sophisticated deep learning model based on a convolutional neural network (CNN) and features extracted from spectrograms. The research focuses on identifying 18 specific isolated words from a dataset of 2,000 audio files collected from native Amazigh speakers in Morocco's Rif region. To accurately represent the speech signal, our system employs spectrograms that plot time on the x-axis and frequency on the y-axis, while indicating the amplitude through the intensity value at a specific position in the spectrogram. For our system architecture, spectrograms act as input to the deep CNNs. We use 1D convolutional neural network structures consisting of eight layers, primarily used for feature learning and recognition. The model extracts discriminative features from spectrogram images and outputs predictions for the eighteen classes. The findings illustrate that the proposed Convolutional Neural Network achieves an impressive accuracy of 94.77%. This highlights the effectiveness of this approach for automatic speech recognition in the Amazigh language, specifically for single-word recognition.

This study addresses the problem of 3D shape retrieval. While this problem is interesting and emerging as the size of 3D object databases grows rapidly, the main two issues the community has to focus on are: computational efficiency of 3D object retrieval and the quality of retrieval results. In this study we deal with the two considerations, especially the first one namely computational efficiency, by proposing a new technique to retrieve efficiently the 3D-objects in the classified databases which contains 3D objects of different categories. This technique can be coupled with any 3D retrieval method. In this study, we use the Clock Matching Bag-of-Features 3D retrieval method proposed by Lian et al. (2010) since it gives the best result comparing with several methods in particular the view based methods. Instead of systematically matching the object-query with all 3D objects of the target database, our approach restricts the pattern matching on a subset of “good candidates” (the most similar to the query). For a database classified in several classes the retrieval will be oriented to the right class that contains similar objects to the query. In this case, the matching process will be not systematically performed with all objects among the database, but only with objects of right class. Our key idea is to represent each class by one representative that will be used to orient the retrieval process to the right class. Experimental results illustrate the efficiency of our approach.