The proliferation of sports-related multimedia content on the internet has presented interesting research challenges for effective visual search and retrieval. The challenges include poor image quality, a wide range of possible camera angles, pose variations of athletes, text deformations on clothing in motion, and occlusions caused by other objects. To overcome these challenges, a new method for identifying text present on the human body in sports-related images. is proposed in this paper. This method for detecting text in images differs from existing approaches that rely on identifying a player's torso, face, and skin. Instead, it uses an integarted episodic learning approach with inductive learning criteria to identify clothing regions within the image. The process involves combining a Residual Network with a Pyramidal Pooling Module to produce a spatial attention map, and the Progressive Scalable Expansion Algorithm is reformed for text detection from these regions. The experimental results on several benchmarks indicate that the suggested method surpasses existing methods in F1-score and precision. Outcomes on sports images from detection of text nature scene datasets such as CTW1500 and RBNR also show that across various inputs, the method that has been suggested is both dependable and efficient.

Clothing detection, residual network, region proposal network, text detection.