Unsupervised methods like deep image prior (DIP) leverage network priors for denoising without labeled data but suffer from slow convergence and overfitting, while deep random projector (DRP) improves efficiency via fixed weights and a random seed yet remains limited by its fully random initialization. In this work, we propose deep fixed projector (DFP), an enhanced DRP-based framework featuring three synergistic improvements: (1) initializing the seed with the noisy image to align optimization with the clean image manifold, (2) using pre-trained clean-to-clean encoder-decoder weights to embed structural priors and accelerate convergence, and (3) introducing inter-inference consistency (IIC), a self-supervised regularization that enforces output stability under input perturbations to suppress noise and reduce overfitting. Experiments show DFP consistently surpasses DIP, DRP, and recent variants in PSNR, SSIM, and LPIPS, while achieving faster convergence and robust denoising quality. Code is available athttps://github.com/Hu-China/Deep-Fixed-Projector.

The deep image prior (DIP) and its variants have demonstrated the ability to address image denoising in an unsupervised manner using only a noisy image as training data, but practical limitations arise due to overfitting in highly overparameterized models and the lack of robustness in the fixed iteration step of early stopping, which fails to adapt to varying noise levels and image contents, thereby affecting denoising effectiveness. In this work, we propose an effective yet fast early stopping metric (ESM) to overcome these limitations when applying DIP models to process synthetic or real noisy images. Specifically, our ESM measures the image quality of the output images generated by the DIP network. We split the output image from each iteration into two sub-images and calculate their distance as an ESM to evaluate image quality. When the ESM stops decreasing over several iterations, we end the training, ensuring near-optimal performance without needing the ground-truth image, thus reducing computational costs and making ESM suitable for application in the denoising of real noisy images.

In this letter, a two-stage noise level estimation (NLE) algorithm that jointly exploited automatic feature extraction and mapping model was proposed. In contrast to existing NLE algorithms using hand-crafted features, we first utilized convolutional neural network-based model to automatically extract the noise level-aware features (NLAFs) in form of feature vector to characterize the distortion degree of a noisy image, i.e., noise level. Then, the NLAF vector was directly mapped to its corresponding noise level via pretrained mapping model, obtaining a fast and reliable NLE algorithm. Extensive experimental results show that the proposed NLE algorithm works well for a wide range of noise levels, showing a good compromise between speed and accuracy.

In this letter, a novel multiple image-based Gaussian noise level estimation (NLE) algorithm for natural images by jointly exploiting the noise level-aware feature extraction and the local means (LM) estimation techniques was proposed. We employed some efficient and powerful noise level-aware features in the form of a feature vector to characterize the noise levels across image contents. Based on this, we adopted LM estimation scheme to estimate the noise level for an image to be estimated by comparing multiple images of similar noise levels in a preconstructed sample database. We had verified the accuracy and efficiency of the proposed NLE algorithm on a large of images from several benchmark databases. Compared with the competing NLE algorithms, our algorithm is superior to them for noise level estimation in terms of both estimation accuracy and execution time.

Unsupervised image segmentation is a fundamental but challenging problem in computer vision. In this paper, we propose a novel unsupervised segmentation algorithm, which could find diverse applications in pattern recognition, particularly in computer vision. The algorithm, named Two-stage Fuzzy c-means Hybrid Approach (TFHA), adaptively clusters image pixels according to their multichannel Gabor responses taken at multiple scales and orientations. In the first stage, the fuzzy c-means (FCM) algorithm is applied for intelligent estimation of centroid number and initialization of cluster centroids, which endows the novel segmentation algorithm with adaptivity. To improve the efficiency of the algorithm, we utilize the Gray Level Co-occurrence Matrix (GLCM) feature extracted at the hyperpixel level instead of the pixel level to estimate centroid number and hyperpixel-cluster memberships, which are used as initialization parameters of the following main clustering stage to reduce the computational cost while keeping the segmentation performance in terms of accuracy close to original one. Then, in the second stage, the FCM algorithm is utilized again at the pixel level to improve the compactness of the clusters forming final homogeneous regions. To examine the performance of the proposed algorithm, extensive experiments were conducted and experimental results show that the proposed algorithm has a very effective segmentation results and computational behavior, decreases the execution time and increases the quality of segmentation results, compared with the state-of-the-art segmentation methods recently proposed in the literature.