En la ?ltima d?cada se han venido desarrollando t?cnicas avanzadas para la eliminaci?n de ruido en el procesamiento de im?genes, cuyos resultados permiten no s?lo una mayor eliminaci?n de ruido, sino la conservaci?n de los bordes. Entre estas t?cnicas se destacan el m?todo del promedio no local (non local means), BDM3D y basados en modelado disperso (sparse modeling) entre otros. Esta ?ltima t?cnica ha despertado gran inter?s debido a los resultados en los cuales se elimina ruido de gran intensidad. Para ellos hace uso de diccionarios que se construyen a medida que el filtro avanza. Esta t?cnica se ha expandido para ser usada en combinaci?n con redes neuronales. En este trabajo, se implement? el algoritmo de suavizaci?n K-SVD. Este algoritmo es basado en modelado disperso. Con el fin de analizar y conocer las bases, caracter?sticas y resultados obtenidos con el m?todo de modelado disperso se compar? con otras t?cnicas de eliminaci?n de ruido, basadas en m?scaras de convoluci?n. Para ello, se analiz? la respuesta del filtro y sus resultados, en cuanto a la eliminaci?n de ruido y conservaci?n de bordes. Para ello se utilizaron im?genes naturales y sint?ticas con diferentes niveles de ruido.In the last decade, advanced techniques have been developed for the elimination of noise in the image processing, whose results allow not only a greater denoise, but the preservation of the edges. These techniques include the non-local average method, BDM3D and based on sparse modeling, among others. This last technique has aroused great interest due to the results in which high-intensity noise is eliminated. For them, it uses dictionaries that are built as the filter progresses. This technique has been expanded to be used in combination with neural networks. In this work, the K-SVD smoothing algorithm was implemented. This algorithm is based on sparse modeling. In order to analyze and know the bases, characteristics and results obtained with the dispersed modeling method, it was compared with other noise elimination techniques, based on convolution masks. For this, the response of the filter and its results were analyzed, in terms of noise elimination and edge conservation. For this purpose, natural and synthetic images with different noise levels were used.