Autonomous systems for detecting and recognizing license plates are leading advancements in computer vision, with extensive uses in traffic control, safety and vehicle movement monitoring. With the availability of large datasets and recent advances in deep learning techniques, computer vision has undergone a revolution that has greatly improved the accuracy of automatic license plate detection and recognition (ALPDR) systems. There are several difficulties when trying to recognize license plates from low-quality photos. Character recognition accuracy is decreased by factors including low quality, poor lighting, diverse plate fonts and variations in sizes and orientations complicate the precise recognition of characters. In this paper, we have proposed an effective approach to recognize Bangla license plate for diverse-quality images such as low light, blurred and low-resolution. At first, YOLOv8 architecture is initially employed for vehicle detection and it is then used once more for license plate image detection and cropping from discovered vehicle photos. After that, we have classified license plate images using CNN into two groups where group I contains low-light images and group II contains blurred and low-resolution images. Then, PIL (Python Imaging Library) autocontrast function and autoencoder approach are utilized to enhance image quality as well as performed some image transformation in order to improve readability and clarity of the classified diverse images. Finally, EasyOCR framework with the string-matching approach is utilized to recognize license plate characters. We have obtained 92.79% optical character recognition (OCR) accuracy in low light, 89.42% accuracy in low-resolution and 73.10% accuracy in blurry photos.