Abstract: Stereoscopic image quality assessment(SIQA) has potential application in evaluating the performance of 3D video system, while the key ingredient in SIQA is to extract visual features that can simulate human brain neural network. In this study, we propose a new full-reference SIQA method based on extreme learning machine(ELM). The proposed method mainly consists of three components. 1) Feature representation for original and distorted stereoscopic images. Using left, right, and cyclopean images as inputs, by mapping the image information to feature space via three-layer ELM, we can obtain the feature representation. 2) Quality vector between the original and distorted stereoscopic images is obtained by measuring the similarity of the feature representation between the original and distorted stereoscopic images at each layer. 3) Based on the derived 12-dimensional quality vectors and the corresponding subjective scores, a regression model is first trained via ELM, and the trained regression model is used to test the quality score at the test stage. Experimental results show that the proposed method is effective on both symmetrical and asymmetrical stereoscopic image databases, and can achieve high consistent alignment with subjective perception.