Timely and accurate information about patients' symptoms is important for clinical decision making such as adjustment of medication. Due to the limitations of self-reported symptom such as pain, we investigated whether facial images can be used for detecting pain level accurately using existing algorithms and infrastructure for cancer patients. For low cost and better pain management solution, we present a smart phone based system for pain expression recognition from facial images. To the best of our knowledge, this is the first study for mobile based chronic pain intensity detection. The proposed algorithms classify faces, represented as a weighted combination of Eigenfaces, using an angular distance, and support vector machines (SVMs). A pain score was assigned to each image by the subject. The study was done in two phases. In the first phase, data were collected as a part of a six month long longitudinal study in Bangladesh. In the second phase, pain images were collected for a cross-sectional study in three different countries: Bangladesh, Nepal and the United States. The study shows that a personalized model for pain assessment performs better for automatic pain assessment and the training set should contain varying levels of pain representing the application scenario.