نویسندگان
دانشگاه بیرجند - دانشکده مهندسی برق و کامپیوتر
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
According to the abundant application of data clustering as an important approach in pattern recognition, many researches such as image clustering have been done in this field. Most of the suggested solutions for image clustering are based on swarm intelligence algorithm. Big amount of input data in these algorithms leads to an excessive amount of computational time in a way that for each member of the population and also for each iteration of the algorithm the cost of clustering should be considered per all the imported data. In 2012, the author proposed an unsupervised algorithm using improved gravitational search algorithm to cluster colored images. According to the suitable performance of SOM neural networks, this paper firstly, tries to perform a primary clustering on all inputs to decrease the amount of input data to the number of output neurons of SOM neural network as input of the proposed algorithm to make final clustering and automatic determining of the number of the image clusters. The lesser amount of input data causes the higher performance of the algorithm. The results show that not only the previous results are relevantly kept in the new approach, but also the fitness value for some images is improved.
کلیدواژهها [English]
)