Optical Character Recognition (OCR) from smartphone images faces challenges due to varying illumination, background textures, and angular distortions. In this paper, we explore an OCR model using a Convolutional Neural Network (CNN) for feature extraction and a Recurrent Neural Network (RNN) for sequence modeling. We compare the performance of Long Short-Term Memory (LSTM) and Gated Recurrent Units (GRU) for sequence modeling. Our study aims to enhance OCR system speed and accuracy by replacing LSTM with GRU. Test results on the ICDAR 2015 dataset show GRU is more accurate and faster than LSTM under the same conditions. The proposed architecture addresses mobile OCR limitations, where processing time and resources are constrained, while maintaining accuracy across different text styles and orientations. Our model improved in handling challenging scenarios including text size, style, rotation, and skewed text in natural scenes.