Optical images are often corrupted by noise, low contrast, uneven illumination and artefacts, which may pose significant challenges to image analysis, particularly for dense fluids. Traditionally, noise removal and contrast enhancement are achieved by global arithmetic operations on the image as a whole, and/or by image convolution with various kernels. However, these methods work under very limited conditions and can compromise detail within the image. Here, we develop a new technique, texture analysis microscopy (TAM), to overcome these challenges based on the method of image correlation. TAM recasts an image by the statistical similarities between a raw image and a template feature (e.g. a Gaussian) that best approximates features in the image. We demonstrate the superiority of TAM by applying it to low-fidelity images under conditions where traditional methods fail or have deteriorative performance, for analyses including structural correlations, particle identification and sizing.