Recently, Indonesian Health Ministry has problems with interpreting health surveillance chart, because they need experts for interpreting the chart. Since it is difficult to provide experts especially in rural area, intelligent system for interpreting the health surveillance chart is necessary. This paper proposed a method to generate sentences in Indonesian language which is extracted from the chart image (e.g. health surveillance chart). The sentences is used to interpret the chart in Indonesian Language. In order to generate sentences, lexicalization based on image features data is proposed to provide some "seed words". These words are used by the NLG system to generate sentences. After obtaining the seed words, the lexical selection is conducted and the main topic of the paragraph is determined. Finally, the main topic was determined and furthermore the sentence in Indonesian Language is ready to be generated. There were some methods for interpreting data such as [1][2][3], but this method could not be used in Indonesian Health Ministry case because most of the chart which should be interpreted are in the form of chart image. Therefore, the proposed method uses NLG based on image extraction and pattern abstraction, such that it can be used for interpreting the chart image.