Provisioning a Pre-Trained Network Model for Extracting Text from Images

Accurate text recognition and detection results for complex action-filled video images are very hard to achieve. To simplify the problem and enhance text detection and recognition capabilities, this work suggests Deep Convolutional Encoder Model (DCEM) model for categorizing text from images with an action focus. This work builds the VGG16 network and ResNet50 to discover the properties of maximally stable extremal regions and acquires the general pixel-distribution level knowledge required to categorize action-oriented video. A multitask convolutional encoder network is utilized to acquire text components which are subsequently introduced to another network model. The method defines text-based classes by combining the outputs of the proposed DCEM model. To demonstrate the efficacy of the proposed strategy, experimentation trials were conducted on text dataset (10 classes of text information). Compared to previous work, the proposed method significantly improves specific text recognition.