According to the latest report from the World Health Orgamization (WHO), malaria remains a disease having a strong negative impact on the African population, especially Ivory Coast population. In view of the health, economic and financial burden caused by this disease, our study was to develop new techniques for malaria diagnosis in Ivory Coast. Our contributions in this thesis concern two major levels: the construction of an optical system and the development of image processing techniques, which combined together allowed the distinction of infected blood cells from uninfected ones.First, following a partnership between the Laboratory of Instrumentation, Image and Spectroscopy (L2IS) and the University of LUND in Sweden, we built a new model of multispectral and multimodal microscope. It is an improvement of the standard optical microscope found in health centers in Ivory Coast. It is also an improvement of the microscope (multispectral and multimodal) that we found arriving at L2IS. The system we built has the particularity of being extensible. It means it can be adapted to the needs of the user both in terms of imaging modalities and in terms of the illumination sources used. We modified and adapted it to malaria diagnosis purpose. For this purpose, we equipped the system with lasers whose wavelengths are 405 nm, 450 nm, 538 nm and 638 nm. The wavelengths chosen are those which are the most discriminating for malaria parasites detection. Our experiments were performed by acquiring multispectral images using the system we built and the one we found at L2IS upon our arrival (it consists of 13 light emitting diodes ranging from 375 nm to 940 nm). We finally obtain multi-component images consisting of the one hand of four (4) spectral planes and on the other hand of thirteen (13).Second, we developed image processing techniques to analyze images produced using the aforementioned multispectral and multimodal microscopes. The works of this part were performed following a partnership between L2IS and LARIS (Laboratoire Angevin de Recherche en Ingénierie des Systèmes) from University of Angers. These works were based on a multi-component texture analysis through Local Binary Pattern (LBP)technique. Resulting from statistical approaches, the choice of Local Binary Pattern (LBP) for the analysis of our images instead of frequency, morphological and structural approaches was adopted because of their simplicity and their robustness in textures classification. Its principle is based on locally describing a texture. Different variants have been developed to make them more robust: on the one hand, those improving the topology of the neighborhood, the sampling of the neighborhood, the thresholding, the quantization. On the other hand, the grouping and encoding of bits and finally the combination of LBP methods with each other or with other methods. We implemented several algorithms to conduct a comparative study between statisticalapproaches and other approaches on the one hand, and between different statistical approaches on the other 17hand. These algorithms were applied to the reference image databases that are Outex, Brodatz and Curet. From comparison tests, it emerges that the statistical approaches denote a higher rate of well-classified than those of other texture analysis approaches. Concerning the comparison tests from statistical approaches between them, the results show that the methods combining several variants of LBP provide better texture classification.Locals Binary Patterns were originally developed for grayscale images. Therefore, they are not straightforward applicable to multispectral images due to the presence of vectors instead of grayscale levels.