In this study, we propose a system called Soft Tactile skIn (STI), which consists of an artificial outer skin and cells. STI can be attached to an existing robot to provide it with flexibility and tactile. The artificial outer skin with sponge structure, which is easy to obtain and process, and the optical sensor is a small photoreflector. To handle the information of multiple cells as a surface, our system generates images from the sensor values and estimates the state using a convolutional neural network, which has shown good performance in the field of image recognition. The system's tactile sensors were able to estimate the contact position with 91.4% accuracy and five contact movements with 81.0% accuracy under specific conditions. STI is expected to expand the scope of Human Robot Interaction (HRI) with soft robots in the future. The contribution of this paper is to propose a method for providing tactile sensors to soft robots based on conventional robots, which combines the features of cell-skin and optical sensors and enables a wide range of surface sensing without spread cells all over a surface.