ICTACT Journal on Image and Video Processing

R. Jayadurga ¹, M. Sathiya ², G.K. Arpana ³

Affiliations
1 Department of Computer Science, Soundarya Institute of Management and Science, India., IN
2 Department of Information Technology, Karpagam Institute of Technology, India., IN
3 Department of Electronics and Communication Engineering, East West College of Engineering, India., IN

Abstract

Most of the time, these classifiers are trained using general-purpose datasets with a lot of classes. Therefore, the performance of these classifiers may not be as good as it could be. Both choosing classifiers based on registrations and dividing them into groups based on the subjects they cover are possible solutions that could lead to better classifier performance. This makes it clear that a classifier division and selection strategy needs for the proposed optimization to work. With the help of this method, the proposed model for feature extraction can choose an appropriate classifier while taking subscription constraints into account. There are subscriptions with the best values of n, and the results of using only n-class classifiers from one domain and ignoring classes from other domains are also given. These are in the same place as the effects of only using n-class classifiers from a certain domain. In this article, these are talked about in the same context as what happens when you only use n-class classifiers from a certain domain. For high-performance use of SAE-based systems, you need to use a classifier selection technique. This method is also needed for the investigation of multimedia events that need the method. To establish the effectiveness of the multimedia event-based system as well as its dependability, we are making use of traditional evaluation methods such as throughput and accuracy. These measures include the following: When compared to the efficiency of the system when using a classifier with a single class, the efficiency of the system diminishes as the number of classes per classifier increases. This is the case regardless of the other measures. This is the situation about both the throughput and the precision of the operation.

Keywords

Multimedia Data, Stacked Auto Encoder, Deep Learning, Classifier.

Full Text

     

M. Sathiya ¹, K. Karunambiga ², G. Sai Kumar ³, S. Chandra Sekaran ⁴

Affiliations
1 1Department of Information Technology, Karpagam Institute of Technology, India., IN
2 Department of Computer Science and Engineering, Karpagam Institute of Technology, India., IN
3 DVR and Dr. HS MIC College of Technology, India., IN
4 Department of Computer Science and Engineering, PSV College of Engineering and Technology, India., IN

Abstract

In recent years, there has been a meteoric rise in the application of deep neural networks for the purpose of iris segmentation. This can be attributed to the extraordinary capacity for learning possessed by the convolution kernels that are utilised by CNNs. Conventional methods have several drawbacks, some of which can be partially compensated for by using CNN-based algorithms, which increase the segmentation precision. On the other hand, the CNN-based iris segmentation approaches that are currently in use typically require a more complex network, which results in an increase in the number of parameters. This is essential to realise a higher degree of precision in the results. CNN-based techniques are effective, they can only be used for a specific application. This makes them inappropriate for general iris segmentation goals, even though they are effective.

Keywords

Ensemble Model, Deep Learning, Iris Segmentation.

Full Text