The multiplexing capability of fluorescence microscopy is vital to study complex intracellular organizations and interactions but limited by the broad fluorescence spectral width. Spectral imaging, together with unmixing techniques, offers potential solutions to discriminate highly overlapping fluorescence signals from multiple targets but struggles to achieve accurate and consistent reference spectra in practical applications. In this study, we present a robust quasi-blind unmixing technique, named Regularization-stabilized Minimization of Mutual Information (Re-MMI), to untangle spectrally and spatially overlapping targets within cells. Re-MMI does not rely on measured spectra during experiments nor does it necessitate accurate reference spectra. Instead, it leverages approximate spectra, e.g., excitation spectra provided by manufacturers, to significantly improve the accuracy of spectral unmixing through mutual information minimization. We validate the concurrent imaging of six fluorophores with substantial spectral overlap, achieving exceptionally low (∼0.5%) crosstalks and high (∼95%) fidelity even against challenging levels of colocalizations, signal-to-noise ratio, and spectral separations. In cell imaging experiments, Re-MMI, when integrated with excitation microscopy, enables high-throughput multitarget imaging and unambiguous identification of up to six different intracellular structures, revealing intricacies and complexity of intracellular conformations and interactions at high spatiotemporal resolution. Together, the capacity to robustly and precisely unmix spectral images without the requirement for accurately measured reference spectra empowers us to visualize complex intracellular organizations and dynamics with a broader spectrum of colors.