In the wake of the burgeoning Internet of Things (IoT) era and the increasing prevalence of image-based applications on mobile platforms, a significant demand for computing resources has been witnessed. While traditional cloud computing has been limited by substantial transmission distances and notable response delays, mobile edge computing, where communication, computation, and storage resources are situated on edge devices, has emerged as a superior alternative. In this context, the challenge of offloading image processing tasks for multiple users, especially considering the collaboration of edge servers under computational and communication resource constraints, is investigated. A primary objective is to strike a balance between energy consumption and task delays, thereby aiming to curtail the total associated costs. The novel framework introduced, termed as Image Collaborative Task Offloading System using Deep Reinforcement Learning (I-CTOS-DRL), is specifically designed for image processing tasks in edge-cloud collaborative scenarios. Through the integration of a set updating mechanism, complications arising from interactions with neighboring edge servers are effectively diminished. Simultaneously, a heuristic algorithm was constructed to identify the most viable servers for task offloading purposes. Building on this foundation, a pioneering methodology for image processing task offloading was devised, leveraging fully connected neural network training. Evaluations conducted extensively indicate that the proposed strategy outperforms established benchmarks in terms of efficiency.