Extracting image features for classification by two-tier genetic programming

Abstract

Image classification is a complex but important task especially in the areas of machine vision and image analysis such as remote sensing and face recognition. One of the challenges in image classification is finding an optimal set of features for a particular task because the choice of features has direct impact on the classification performance. However the goodness of a feature is highly problem dependent and often domain knowledge is required. To address these issues we introduce a Genetic Programming (GP) based image classification method, Two-Tier GP, which directly operates on raw pixels rather than features. The first tier in a classifier is for automatically defining features based on raw image input, while the second tier makes decision. Compared to conventional feature based image classification methods, Two-Tier GP achieved better accuracies on a range of different tasks. Furthermore by using the features defined by the first tier of these Two-Tier GP classifiers, conventional classification methods obtained higher accuracies than classifying on manually designed features. Analysis on evolved Two-Tier image classifiers shows that there are genuine features captured in the programs and the mechanism of achieving high accuracy can be revealed. The Two-Tier GP method has clear advantages in image classification, such as high accuracy, good interpretability and the removal of explicit feature extraction process. © 2012 IEEE.

Genetic Programming for Automatic Global and Local Feature Extraction to Image Classification

“…1) GP methods: The seven GP methods are MLGP [28], GP-RDP [29], 2TGP [26], 3TGP [25], DIF+GP [7], Hist+GP, and uLBP+GP. The MLGP, GP-RDP, 2TGP, and 3TGP methods take an image as input and return a high-level image feature for binary image classification.…”

“…The extracted features are based on the automatically detected regions, which are local features. Atkins et al [25] developed a multi-tier GP approach (simplified as 3TGP in [26]) to achieving automatic feature extraction and image classification. In this method, three tiers including an image filtering tier, an aggregation tier and a classification tier, were designed where each tier targeted a subtask.…”

“…Later, the 3TGP approach was changed by Al-Sahaf et al [26] who proposed a two-tier GP (2TGP) approach to image feature extraction and classification using raw images as input. The representation of 3TGP was simplified in 2TGP where only the aggregation tier and the classification tier were employed.…”

“…GP has been applied to automatic feature extraction and image classification based on raw images [6,18,[22][23][24][25][26][27][28][29]. In [18,[25][26][27], GP is employed to region detection, feature extraction, feature construction, and image classification. These methods capture general local information/features, i.e.…”

“…However, the work of GP on feature extraction only focus on extracting one type of image features, i.e. local features [18,[25][26][27][28][29] or global features [6,[22][23][24], which are often not effective and efficient for complex image classification. In addition, existing image descriptors such as HOG, SIFT and LBP are the well-known effective methods for feature extraction, but they are seldom cooperated with GP to automatically extract more adaptive and informative features for image classification.…”