This study presents a novel approach to image representation, utilizing wavelet transforms to compress image information into a compact latent space. Wavelet transforms offer a multi-resolution analysis, decomposing images into different frequency components at different resolution scales, emphasizing the spatial and frequency attributes. By leveraging the hierarchical structure of wavelet coefficients, we construct a latent space representation preserving essential features while reducing dimensionality. Experimental evaluations on benchmark datasets demonstrate competitive performance in tasks such as compression and classification compared to traditional deep learning approaches. Wavelet-based representation offers promise for addressing challenges in high-dimensional data while retaining crucial image information for diverse processing tasks.