An Efficient CNN-LSTM Based Framework for Improved Image Captioning

An image caption serves as a written explanation of a picture and finds widespread application in programs requiring automatic extraction of information from images. In the today’s world of vision innovation, the generation of descriptive textual representations for images, known as image captioning, holds significant importance across various applications. This paper presents an efficient framework for image captioning that make a fusion of of Convolutional Neural Networks (CNNs) and Long Short-Term Memory networks (LSTMs) for providing more precise and enriched captions. The proposed model utilizes CNNs to extract rich visual features from images, which are then fed into LSTMs for sequential caption generation. By integrating advanced feature extraction with robust sequence modeling, the framework aims to improve the standard generated captions, capturing both the key objects and contextual details within the image. The proposed framework is introducing enhancements to LSTM architecture for improved efficiency, with a comparative evaluation between LSTM and Gated Recurrent Unit (GRU) methods. Extensive experiments were conducted using Flickr8k dataset. Through rigorous testing utilizing BLEU Metrics, LSTM emerges as the superior performer, achieving an impressive 82% efficiency rate.