Recently, many classes of objects can be efficiently detected by the way of machine learning techniques. In practice, boosting techniques are among the most widely used machine learning for various reasons. This is mainly due to low false positive rate of the cascade structure offering the possibility to be trained by different classes of object. However, it is especially used for face detection since it is the most popular sub-problem within object detection. The challenges of Adaboost based face detector include the selection of the most relevant features from a large feature set which are considered as weak classifiers. In many scenarios, however, selection of features based on lowering classification errors leads to computation complexity and excess of memory use. In this work, we propose a new method to train an effective detector by discarding redundant weak classifiers while achieving the pre-determined learning objective. To achieve this, on the one hand, we modify AdaBoost training so that the feature selection process is not based any more on the weak learner’s training error. This is by incorporating the Genetic Algorithm (GA) on the training process. On the other hand, we make use of the Joint Integral Histogram in order to extract more powerful features. Experimental performance on human faces show that our proposed method requires smaller number of weak classifiers than the conventional learning algorithm, resulting in higher learning and faster classification rates. So, our method outperforms significantly state-of-the-art cascade methods in terms of detection rate and false positive rate and especially in reducing the number of weak classifiers per stage.
Wavelet performances differ from one application to another and from one database to another. In this case, one can try to find out for each application the appropriate wavelet transform which results in better performances and consumes minimal resources once implemented on an FPGA platform. Accordingly, we use a generic lifting wavelet transform with p0 and q parameters. Thus, we train the optimisation process with a multi-objective genetic algorithm for optimising wavelet transform with respect to specific applications. Optimised criteria are related to wavelet regularity and the undertaken application performances. We consider pattern recognition and image compression applications processed respectively on ORL database and a fingerprint database. Compared with DB 4 and DB 9/7, the improvement of the criterion related to the application reaches 17 % for pattern recognition application and preserves, approximately, the same values for still image compression in addition to the minimisation of the hardware cost.
A key challenge in computer vision applications is detecting objects in an image which is a non-trivial problem. One of the better performing proposed algorithms falls within the Viola and Jones framework. They make use of Adaboost for training a cascade of classifiers. The challenges of Adaboost-based face detector include the selection of the most relevant features which are considered as weak classifiers. However, selection of features based on lowering classification error leads to high computation complexity. To overcome this limitation, a novel genetic Adaboost is proposed in our work. In the same context of optimisation, a selection method based on Pareto concept of the most relevant features referred to as dominant features is proposed. This optimisation allows to reduce the initial feature space by 28%. Moreover, we notice that dominant features with genetic Adaboost further improve the performance of genetic Adaboost, reducing the total number of features by 20%.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.