Multiple imputation on categorical (dummy coded) data


I have a question regarding multiple imputation for categorical variables, and hope very much that someone can answer me.
I am using SPSS 20.0.

So, I have a few categorical variables with several levels (for example, color, with blue, red, green, and yellow as levels) and have missings on some of the levels. I would like to perform a linear regression once the missings are imputated.
For first regression analyses on my complete cases, I dummy-coded the categorical variables. However, I am concerned that when I impute, SPSS will assign a "1" to more than one color, thereby ruining the dummy-coding. But imputing the original variable with four levels also doesn't work because I need binary variables for the regression.

From what I learned by googling, most sites advise to make a binary variable of the categorial variable, but unfortunately, that isn't an option for me.

Could someone help me and explain what I should do?

Thanks so much