A Semantic Image Classifier Based on Hierarchical Fuzzy Association Rule Mining

Submitted to MULTIMEDIA TOOLS AND APPLICATION, 2012

Abolfazl Tazaree^a, Amir-Masud Eftekhari-Moghadam^b, Saeedeh Sajjadi-Ghaem-Maghami^c

Faculty of Electrical, Computer and IT Engineering, Azad University, Qazvin Branch, Qazvin, Iran.

^a tazaree@gmail.com, ^b eftekhari@qiau.ac.ir, ^c s.ghaemmaghami@qiau.ac.ir

Abstract One major challenge in the content-based information retrieval and machine learning is to build the so-called "semantic classifier" which effectively and efficiently classifies semantic concepts in a large database. This paper deals with the semantic image classifier based on hierarchical Fuzzy Association Rules (FARs) mining in the image database. Intuitively, an association rule is a unique and significant combination of image features and a semantic concept, which determines the grade of correlation between features and concept. The main idea behind this approach is that any image visual concept has some associated features, which there are strong correlations between the concept and its corresponding features. Regardless of the semantic gap, an image concept appears when the corresponding features emerge in an image and vice versa. Specially, paper contribution is to propose a novel Fuzzy Association Rule for improving traditional association rules. Moreover, we are concerned that establishing a hierarchical fuzzy rule base in training phase and setup corresponding fuzzy inference engine to classify images in testing phase. Presented approach is independent of image segmentation also can apply on multi-label images. Experimental results on a database of 6000 general-purpose images demonstrate the superiority of the proposed algorithm.

Keywords Semantic image classifier, Hierarchical fuzzy classification, Fuzzy Association Rule, fuzzy expert systems.

Download full paper [PDF] [xps]

Download UCI dataset

Download image database:

Layer 1	Layer 2	Layer 3	Layer 4	Concept ID
Artificial	Indoor	Kitchen	Desk	1
			Plate	2
			Cup	3
			Cabinet	4
			Fork	5
			Spoon	6
			Glass	7
		Computer	Mouse	8
			Keyboard	9
			Monitor	10
		Toy	Car	11
			House	12
			Dolly	13
		Seat	Sofa	14
			Chair	15
	Out door	House	Ancient	16
			Modern	17
			Jackal	18
		Vehicle	Bicycle	19
			Airplane	20
			Bus	21
			Train	22
Nature	Animal	Bird	Duck	23
			Butterfly	24
			Stork	25
			Eagle	26
		Non-bird	Bear	27
			Fox	28
			Fish	29
			Dinosaur	30
			Monkey	31
			Cat	32
			Cow	33
			Camel	34
			Horse	35
			Elephant	36
	Plant	Fruit	Date	37
			Apple	38
			Tomato	39
			Pear	40
		Flower	Rose	41
			Kaktos	42
			Mushroom	43
	Landscape	Waterscape	Sea	44
			Waterfall	45
		Mainland	Sunset	46
			Mountain	47
			Snow	48