This paper proposes a two-class classification model for image segmentation, using Gestalt grouping cues as features. Human segmented images are used as positive examples, while negative examples are constructed by randomly matching human segmentations with different images. The preprocessing stage involves oversegmenting images into superpixels, and features derived from Gestalt cues such as contour, texture, brightness, and good continuation are defined. These features are evaluated using information-theoretic measures, and a logistic regression classifier is trained to combine them. The segmentation problem is then formulated as an optimization problem over the space of segmentations. A simple random search algorithm is used to find good segmentations, and the results are evaluated using precision-recall analysis. The model is shown to be effective in segmenting a wide range of images. The key contributions include the use of Gestalt cues in a learning framework, the normalization of features, and the formulation of segmentation as a classification problem. The results demonstrate that the model performs well in capturing the information contained in the grouping cues, and that a linear classifier is sufficient for this task. The paper also discusses related works and concludes that the proposed approach is effective in segmenting natural images.This paper proposes a two-class classification model for image segmentation, using Gestalt grouping cues as features. Human segmented images are used as positive examples, while negative examples are constructed by randomly matching human segmentations with different images. The preprocessing stage involves oversegmenting images into superpixels, and features derived from Gestalt cues such as contour, texture, brightness, and good continuation are defined. These features are evaluated using information-theoretic measures, and a logistic regression classifier is trained to combine them. The segmentation problem is then formulated as an optimization problem over the space of segmentations. A simple random search algorithm is used to find good segmentations, and the results are evaluated using precision-recall analysis. The model is shown to be effective in segmenting a wide range of images. The key contributions include the use of Gestalt cues in a learning framework, the normalization of features, and the formulation of segmentation as a classification problem. The results demonstrate that the model performs well in capturing the information contained in the grouping cues, and that a linear classifier is sufficient for this task. The paper also discusses related works and concludes that the proposed approach is effective in segmenting natural images.