Alzheimer is a memory depletion disease, which is widely recognized as dementia. The research on early detection of dementia has received huge interest among the researchers to help in reducing mortality rates of Alzheimer's patients. In recent years in the medical field, the deep learning techniques play an important role in computer aided diagnosis. In this research, the automatic recognition of Alzheimer Disease (AD) based on the Magnetic Resonance Imaging (MRI) is accomplished by implementing an unsupervised classification technique named Deep Neural Network (DNN) with the rectified Adam optimizer. At first, Histogram of Oriented Gradients (HOG) is utilized to extract the feature values from the brain images, which were acquired from National Institute of Mental Health and Neurosciences (NIMHANS) and Alzheimer disease Neuroimaging Initiative (ADNI) datasets. Next, the extracted features were given as the input to DNN with the rectified Adam optimizer to distinguish the healthy, AD and Mild Cognitive Impairment (MCI) patients. The experimental results have revealed that the HOG-DNN with the rectified Adam optimizer has achieved better performance in AD recognition and showed 16% enhancement in classification accuracy compared to other existing work; Landmark based features with support vector machine classifier.