This paper presents a novel steganalysis method for detecting steganographic methods that embed messages in the spatial domain by adding low-amplitude independent stego signals, such as LSB matching. The method uses first-order and second-order Markov chains to model pixel differences and employs support vector machines (SVMs) for steganalysis. The accuracy of the steganalyzer is evaluated on LSB matching and four databases, showing superior performance compared to prior methods. The method addresses the curse of dimensionality by using a feature selection algorithm, demonstrating that it does not affect results. The steganalyzer is based on the subtractive pixel adjacency matrix (SPAM) model, which captures pixel dependencies by analyzing differences between adjacent pixels. The SPAM features are derived from transition probability matrices of Markov chains, with the second-order model showing better performance than the first-order. The method is tested on four databases, including CAMERA, BOWS2, NRCS, and JPEG85, and compared with existing detectors like WAM and ALE. Results show that the SPAM steganalyzer outperforms these methods, with the second-order model providing the best accuracy. The method uses a Gaussian kernel SVM for classification and evaluates performance using the minimal average decision error. The results indicate that the decision boundary between cover and stego images is nonlinear, and the second-order SPAM model is more effective. The study also investigates the curse of dimensionality, finding that it does not occur in the experiments, even with high-dimensional feature sets. The SPAM features are shown to be robust across different cover sources and provide stable results. The paper concludes that the proposed method offers a significant improvement in steganalysis accuracy and is a promising approach for detecting steganographic methods in the spatial domain.This paper presents a novel steganalysis method for detecting steganographic methods that embed messages in the spatial domain by adding low-amplitude independent stego signals, such as LSB matching. The method uses first-order and second-order Markov chains to model pixel differences and employs support vector machines (SVMs) for steganalysis. The accuracy of the steganalyzer is evaluated on LSB matching and four databases, showing superior performance compared to prior methods. The method addresses the curse of dimensionality by using a feature selection algorithm, demonstrating that it does not affect results. The steganalyzer is based on the subtractive pixel adjacency matrix (SPAM) model, which captures pixel dependencies by analyzing differences between adjacent pixels. The SPAM features are derived from transition probability matrices of Markov chains, with the second-order model showing better performance than the first-order. The method is tested on four databases, including CAMERA, BOWS2, NRCS, and JPEG85, and compared with existing detectors like WAM and ALE. Results show that the SPAM steganalyzer outperforms these methods, with the second-order model providing the best accuracy. The method uses a Gaussian kernel SVM for classification and evaluates performance using the minimal average decision error. The results indicate that the decision boundary between cover and stego images is nonlinear, and the second-order SPAM model is more effective. The study also investigates the curse of dimensionality, finding that it does not occur in the experiments, even with high-dimensional feature sets. The SPAM features are shown to be robust across different cover sources and provide stable results. The paper concludes that the proposed method offers a significant improvement in steganalysis accuracy and is a promising approach for detecting steganographic methods in the spatial domain.