Improved Image Fusion Technique Using Convolutional Neural Networks and The Hybrid PCA-Guided Filter

Abstract

In order to better accurately forecast the diagnostic features, medical image fusion attempts to combine multi-focus and multimodal medical data into an original image. The outcomes of deep learning-based image processing can be visually beautiful. The incomprehensibility of the outcomes in Medical Imaging is a critical topic. This paper offers a feature-level multi-focus, multi-exposure, and multimodal image fusion using a hybrid layer of Principal Component Analysis (PCA) and Guided Filter (GF) to maximize the anatomical details and eliminate significant noise and artefacts. The proposed method utilizes a Convolutional Neural Network (CNN) based network for feature extraction. The original image is initially decomposed using Principal Component Analysis (PCA). A PCA decreases its dimensionality while preserving all of the essential information in the picture in the first stage and produces a revised weight map. A Guided Filter is used at the PCA output to uphold the edges and further augment the features, reducing the ringing and blurring effects. In the third step, a pre-trained CNN network creates a new weight map by extracting critical characteristics from pictures in the input dataset. The output feature map combines the weight maps produced by the GF and CNN, which is further fused with the reference picture to create the fused image output. The contribution of developed method:• To improvise image quality features by removing noise, ringing and blurring.• Increase the quality of the image by using the hybrid mechanism for extracting more underlying critical features of images [1]. The estimation is based on three multimodal imaging datasets, including CT-MRI, MRI-PET, and MRI-SPECT. Furthermore, the proposed method excels in existing state-of-the-art techniques in terms of fusion quality. .